Regional ecological environment monitoring method and system based on remote sensing image processing

Abstract

The invention relates to the technical field of ecological environment monitoring, and discloses a regional ecological environment monitoring method and system based on remote sensing image processing, and the method comprises the following steps: collecting remote sensing image data and climate factor data of a target region; performing spatial gridding and standardization processing on the data; establishing a nonlinear regression model to analyze a relation between climate factors and vegetation indexes; extracting spatio-temporal topological characteristics of the climate factor data and identifying a change mode; long-term prediction of regional vegetation productivity is carried out through time sequence model training; the regional ecological environment is analyzed and evaluated based on the prediction result, and a scientific basis is provided for ecological environment monitoring; the system comprises a data acquisition module, a data processing module, a nonlinear regression modeling module, a spatio-temporal topology analysis module, a time sequence prediction module and an ecological evaluation module. According to the method, the problems of small data coverage, low prediction precision and insufficient capture of nonlinear influence of climate change in the existing method are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of ecological environment monitoring, and in particular to a regional ecological environment monitoring method and system based on remote sensing image processing. Background Art

[0002] With the increasing severity of global climate change and environmental pollution problems, the monitoring and protection of the ecological environment have become important issues to be solved globally. Traditional ecological environment monitoring methods mainly rely on ground observations and manual sampling. These methods are limited by the uneven spatio-temporal distribution of data collection and the influence of human intervention, resulting in constraints on the timeliness and accuracy of ecological monitoring. Ground monitoring cannot cover large areas, and the delay in periodic data updates often makes it difficult to respond to and evaluate environmental changes in real time. Therefore, how to improve the coverage, timeliness, and accuracy of ecological environment monitoring has become one of the main challenges in current research and applications.

[0003] With the development of remote sensing technology, environmental monitoring methods based on remote sensing images have gradually become the mainstream. These methods can obtain surface data over a large range and in real time, especially in areas such as vegetation cover, land use, and climate change, where significant progress has been made. However, existing remote sensing image processing technologies mostly focus on single-dimensional data analysis and often ignore the comprehensive impact of multiple climate factors on vegetation productivity. Traditional remote sensing data processing mostly uses linear models or simple statistical analysis methods, which makes it difficult to handle complex climate change patterns and dynamic feedback of ecosystems in practical applications.

[0004] In addition, existing ecological environment assessment models usually have limitations in spatial and temporal scales. Most technologies only perform simple statistical analysis on climate factors when analyzing the regional ecological state and fail to deeply explore the complex change patterns of climate factors in different temporal and spatial dimensions. This makes the results of ecological risk assessment often lack foresight and it is difficult to accurately predict future ecological environment evolution trends.

[0005] Therefore, the defects of traditional methods are mainly manifested in the following aspects: First, the single nature of data sources and the non-uniformity of spatial and temporal scales lead to limited coverage of ecological environment monitoring; second, the simplified processing of models fails to fully capture the non-linear relationship between climate factors and vegetation growth; third, there are deficiencies in the analysis and prediction of spatio-temporal changes and dynamics, and it is impossible to effectively respond to the increasingly changing ecological environment trends.

[0006] Therefore, the present invention proposes a regional ecological environment monitoring method and system based on remote sensing image processing to solve the deficiencies of the existing technology. Summary of the Invention

[0007] In view of the deficiencies of the prior art, the present invention provides a method and system for regional ecological environment monitoring based on remote sensing image processing, which solves the problems of limited coverage of ecological environment monitoring data, low prediction accuracy, and insufficient capture of non-linear impacts of climate change in the prior art.

[0008] To achieve the above objectives, the present invention is realized through the following technical solutions: A method for regional ecological environment monitoring based on remote sensing image processing, comprising the following steps:

[0009] Collect remote sensing image data of the target area and extract vegetation indices;

[0010] Obtain climate factor data of the target area;

[0011] Perform spatial gridding and standardization processing on the remote sensing image data, vegetation indices, and climate factor data;

[0012] Establish a non-linear regression model based on the climate factor data and vegetation index data;

[0013] Extract topological feature data from the climate factor data based on the spatio-temporal topological data analysis method to identify the spatio-temporal change patterns of climate factors;

[0014] Use a time series model to train the climate factor data and topological feature data, so as to perform long-term prediction of regional vegetation productivity changes and output prediction results;

[0015] Conduct regional ecological environment analysis and evaluation based on the prediction results to provide a scientific basis for ecological environment monitoring.

[0016] The present invention also provides a regional ecological environment monitoring system based on remote sensing image processing, comprising:

[0017] A data acquisition module for collecting remote sensing image data of the target area and extracting vegetation indices, and obtaining climate factor data of the target area, where the climate factor data includes at least temperature, precipitation, and sunshine;

[0018] A data processing module for performing spatial gridding and standardization processing on the remote sensing image data, vegetation indices, and climate factor data to generate a raster data set with a unified spatial scale;

[0019] A non-linear regression modeling module for constructing a non-linear regression model based on the climate factor data and vegetation index data, and the model describes the relationship between vegetation productivity and climate factors through a regression equation including quadratic terms and interaction terms of climate factors;

[0020] A spatio-temporal topological analysis module for extracting topological feature data from climate factor data and identifying its spatio-temporal change patterns;

[0021] A time series prediction module, which is used to train climate factor data and topological feature data through a long short-term memory network to generate long-term prediction results of regional vegetation productivity;

[0022] An ecological assessment module, which is used to analyze the change trend of vegetation productivity based on the prediction results, divide ecological risk areas, and output a visual report and management suggestions.

[0023] The present invention provides a method and system for regional ecological environment monitoring based on remote sensing image processing. It has the following beneficial effects:

[0024] 1. The present invention adopts a multi-source data fusion technology based on remote sensing images and combines climate factor data for rasterization processing at a unified spatial scale, achieving efficient and accurate regional ecological monitoring effects. Compared with the traditional method that only relies on a single data source in the prior art, the present invention solves the data island phenomenon by comprehensively analyzing multi-dimensional data, improving the adaptability and accuracy of the system in complex environments.

[0025] 2. The present invention constructs a non-linear regression model to accurately describe the non-linear relationship between vegetation productivity and climate factors, achieving higher prediction accuracy. Compared with the scheme that adopts a linear model in the prior art, the regression model of the present invention can capture the complex impact of climate change on vegetation, solving the deficiency that the traditional method fails to fully explore the complexity of climate factors.

[0026] 3. The present invention introduces a spatio-temporal topology analysis module to deeply analyze the spatio-temporal distribution characteristics of climate factors, achieving accurate identification of the trend of regional ecological changes. Compared with the processing method that simply statistically analyzes climate factors in the prior art, the present invention solves the defect that the traditional method ignores spatio-temporal dynamic changes by extracting and analyzing topological features, improving the prediction ability of the model.

[0027] 4. The present invention trains climate factor and topological feature data through a long short-term memory network (LSTM) to realize long-term prediction of vegetation productivity, achieving a forward-looking ecological environment monitoring effect. Compared with the technical scheme that only relies on short-term data prediction in the prior art, the present invention can accurately capture the long-term ecological change trend, solve the problem that the existing method cannot provide long-term early warning, and provide more reliable decision-making support for ecological protection. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 is a flowchart of the method of the present invention;

[0029] Figure 2 is a system architecture diagram of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0030] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0031] Please refer to the attached Figure 1 , the embodiment of the present invention provides a method for monitoring regional ecological environment based on remote sensing image processing, including the following steps:

[0032] S1. Collect remote sensing image data of the target area and extract the vegetation index;

[0033] For S1, through remote sensing data collection and vegetation index extraction, basic data support is provided for subsequent ecological environment analysis. The steps of collecting remote sensing image data of the target area and extracting the vegetation index need to ensure data quality and spatial consistency.

[0034] In this embodiment, the steps of collecting remote sensing image data of the target area and extracting the vegetation index are implemented in the following manner:

[0035] Remote sensing data source and acquisition control:

[0036] Specifically, when obtaining remote sensing image data of the target area from ground or aerial photography equipment, it is necessary to select an appropriate data source according to the monitoring scale. For example, in the embodiments of the upper and middle reaches of the Yellow River Basin, for gully areas, a drone-mounted multispectral camera (band range 400-900nm) is used to collect sub-meter resolution data, while for large-scale monitoring, 10m resolution multispectral images of Sentinel-2 satellites are used.

[0037] As an option, when there is cloud cover interference in the monitoring area, the time series image fusion technology is used to merge the effective pixels of multiple images within the same quarter to ensure the continuity of vegetation index extraction.

[0038] Radiometric correction and atmospheric correction:

[0039] When preprocessing the obtained remote sensing image data, different calibration parameters are required for radiometric correction according to the sensor type. For example, for Landsat images, the radiance scaling coefficients (RADIANCE_MULT_BAND_x and RADIANCE_ADD_BAND_x) in the header file are used to convert the original DN value into a radiance value:

[0040] L λ = RADIANCE_MULT × DN + RADIANCE_ADD;

[0041] Among them, L λ represents the radiance value at the entrance pupil of the sensor; DN (Digital Number) is the original pixel gray value of the remote sensing image, dimensionless, and its value range is determined by the quantization bit number of the sensor (for example, the DN range of a 12-bit sensor is 0 - 4095); RADIANCE_MULT is the radiation scaling coefficient, used to linearly map the DN value to the physical radiation dimension; RADIANCE_ADD is the radiation offset coefficient, used to correct the dark current or substrate radiation value of the sensor.

[0042] For atmospheric correction, the 6S model or the ENVIFLAASH module is used. By inputting parameters such as aerosol optical thickness and water vapor content, the influence of atmospheric scattering is eliminated. In the example of the Yellow River Basin, ground-measured reflectance data is additionally introduced to verify the correction accuracy, ensuring that the vegetation index error is controlled within ±0.05.

[0043] Geometric correction and spatial registration:

[0044] The geometric correction uses a quadratic polynomial model. The control points are selected from the regional digital elevation model (DEM) and the field measurement coordinates, and the residual tolerance is 0.5 pixels. Taking the Sentinel-2 image as an example, the spatial offset after correction from the Landsat image does not exceed 10m.

[0045] Vegetation index calculation and standardized output:

[0046] Select data from the red light band (for example, LandsatBand4, central wavelength 650nm) and the near-infrared band (for example, LandsatBand5, central wavelength 860nm), and characterize the vegetation coverage through the normalized difference vegetation index (NDVI):

[0047]

[0048] Among them, NIR is the reflectance of the near-infrared band; Red is the reflectance of the red light band.

[0049] In a possible implementation, when there is a mixture of high vegetation coverage and soil background in the monitoring area, the enhanced vegetation index (EVI) is used as a supplement:

[0050]

[0051] Among them, NIR is the reflectance of the near-infrared band (dimensionless); Red is the reflectance of the red light band (dimensionless); Blue is the surface reflectance of the blue light band (dimensionless); G is the gain factor (dimensionless); C1 is the atmospheric correction coefficient of the red light band (dimensionless); C2 is the atmospheric correction coefficient of the blue light band (dimensionless); L is the background adjustment term (dimensionless).

[0052] When outputting the calculated vegetation index data, it is converted into standardized raster data in GeoTIFF format. The spatial reference system is unified as the WGS84 UTM projection, and the pixel value range is mapped to 0 - 255 (8-bit quantization), which is compatible with the parsing of mainstream GIS platforms.

[0053] Generally, when extracting the vegetation index, temporal optimization needs to be combined with the phenological cycle. For example, in the Yellow River Basin implementation, images during the peak vegetation growth season from July to September are preferentially selected to avoid the interference of winter snow cover and spring bare land surface.

[0054] As a possible implementation method, when there is a mixed pixel problem, the linear spectral unmixing technique is adopted to decompose the proportion of vegetation, bare soil, and water endmembers, and the weighted value of the endmember NDVI is used to improve the accuracy of index calculation.

[0055] Through multi-source data fusion, radiation - geometric joint correction, and optimized calculation of vegetation indices, high-precision and highly consistent extraction of vegetation cover information is achieved, providing a reliable data basis for subsequent ecological environment monitoring.

[0056] S2. Obtain the climate factor data of the target area;

[0057] For S2, by fusing multi-source meteorological data and spatial reconstruction technology, climate factor raster data that is spatially matched with the vegetation index is generated to provide standardized input for subsequent modeling. The spatio-temporal resolution consistency and data integrity of climate factors directly affect the reliability of vegetation - climate response analysis. It is necessary to design adaptable data acquisition and reconstruction strategies for challenges such as uneven distribution of meteorological stations and differences in remote sensing meteorological products.

[0058] In this embodiment, the step of obtaining the climate factor data of the target area is implemented in the following way:

[0059] Multi-source data collaborative acquisition and quality control:

[0060] Specifically, temperature, precipitation, and sunshine duration data are obtained through ground meteorological stations, meteorological remote sensing data (such as MODIS land surface temperature products), and historical meteorological databases (such as CRUTS). In the Yellow River Basin implementation, data from 85 ground stations of the National Meteorological Administration and CMORPH satellite precipitation products are integrated, covering the period from 2000 to 2023 in terms of time and the entire upper and middle reaches in terms of space.

[0061] As an option, when the density of meteorological stations in the monitoring area is less than 1 station per 10,000 square kilometers, ERA5 reanalysis data is introduced to fill the spatial blind area, and its accuracy is evaluated using the cross-validation method (with station data as the benchmark), and low-quality grids with root mean square error (RMSE) > 2 times the observational standard deviation are excluded.

[0062] Data Outlier Detection and Missing Value Reconstruction:

[0063] When cleaning climate factor data, the 3σ principle is combined with climatological thresholds for outlier filtering. For example, records of the daily average temperature in July in the Yellow River Basin that are >40°C or <10°C are considered abnormal, and the calculation formula is as follows:

[0064] X i <μ - 3σ or X i >μ + 3σ;

[0065] Among them, X i is the single-station data value; μ is the regional synchronous climate mean, and σ is the standard deviation.

[0066] The spatio-temporal KNN algorithm is used to fill in the missing data: Select the 5 nearest spatial neighboring stations and the data for the previous / next 5 days, and perform weighted average according to the inverse distance:

[0067]

[0068] Among them, X 填补 is the filled value of the missing data at the target station / location; X j is the observed value at the j-th neighboring station / location during the same time period; w j is the spatio-temporal weight of the j-th neighboring station / location; d j is the spatial distance between the target station and the j-th neighboring station; Δt j is the time difference between the target time period and the data time period of the j-th neighboring station; n is the number of valid neighboring stations participating in the calculation.

[0069] Spatial Interpolation and Rasterization Processing:

[0070] The spatial interpolation method is used to convert the station data into a raster with a resolution of 30m, which is strictly aligned with the vegetation index data in step S1. In a possible implementation, in complex terrain areas (such as the gully area of the Loess Plateau), the Co-Kriging method is used, and elevation is introduced as a covariate:

[0071] Z(s) = β0 + β1H(s) + δ(s);

[0072] Among them, Z(s) is the predicted value of the main variable at location s, such as temperature; β0 is the intercept term, that is, the constant term of the regression model; β1 is the regression coefficient of the covariate H(s); H(s) is the covariate value at location s; δ(s) is the residual term of spatial autocorrelation, usually assumed to have a certain spatial structure, such as a Gaussian process. The interpolation result is cross-validated, and it is required that the temperature RMSE ≤ 1.5°C and the precipitation RMSE ≤ 20mm / month.

[0073] Data Standardization and Output Control:

[0074] Normalize the interpolated raster data to the range of [0, 1]. The formula is as follows:

[0075]

[0076] Where X min is the minimum value of the dataset; X max is the maximum value of the dataset; X is the value of the original data point; X norm is the result after normalization.

[0077] The output is in GeoTIFF format, with the spatial reference consistent with the vegetation index data (WGS84 UTM Zone 48N). The single-layer data file is named according to "climate factor_year_month" for easy calling of the time series model.

[0078] Extended technical solution:

[0079] Generally, the sunshine duration data needs to be corrected for topographic shading. For example, in the embodiment of the Yellow River Basin, based on the 30m DEM data, calculate the topographic shadow coefficient α (0 - 1). The correction formula is:

[0080] S 校正 = S 原始 ×(1 + k(1 - α));

[0081] Where S 原始 is the original signal, data or measurement value; α is the reliability coefficient or confidence level, indicating the degree of trust in the original data; k is the gain compensation coefficient, controlling the correction amplitude; S 校正 is the corrected output value.

[0082] S3. Perform spatial gridding and standardization processing on the remote sensing image data, vegetation index, and climate factor data;

[0083] For S3, in this step, for the remote sensing image data, vegetation index, and climate factor data, through spatial gridding and standardization processing, the multi-source data are unified on the same spatial grid, enabling direct comparison and calculation in subsequent analysis. Specifically, spatial gridding and standardization not only ensure the spatial consistency between different data sources but also provide high-quality and standardized data input for subsequent modeling, analysis, and prediction. The key to this step lies in how to accurately process data from different sources to maximize the spatial and numerical consistency.

[0084] In this embodiment, the step of performing spatial gridding and standardization processing on the remote sensing image data, vegetation index, and climate factor data is implemented in the following manner:

[0085] Unify the spatial resolution and coordinate reference system

[0086] Specifically, first, it is necessary to ensure the consistency of remote sensing image data, vegetation index data, and climate factor data in terms of spatial resolution and coordinate reference system. Generally, remote sensing image data has a relatively high spatial resolution (e.g., 30m or higher), and vegetation index data (such as NDVI) usually also presents at a similar spatial resolution. While climate factor data (such as temperature, precipitation, etc.) often comes from meteorological stations, satellite remote sensing, or historical meteorological data, and its spatial resolution may be relatively low. To achieve spatial alignment between the data, all data needs to be unified to the same spatial resolution and coordinate reference system. For example, in this embodiment, the spatial resolution is unified to 30m, and the coordinate reference system uses the WGS84 coordinate system. The specific approach includes using resampling techniques to upsample the low-resolution climate factor data to the target resolution. The resampling method can adopt bilinear interpolation, and the calculation formula is as follows:

[0087]

[0088] Where X new is the new interpolation result at the target position, that is, the interpolation value at the resampled target grid cell;

[0089] X i,j is the data value of the adjacent original grid point; w i,j is the interpolation weight, which is usually inversely proportional to the spatial distance between the adjacent point and the target position, indicating the contribution degree of each adjacent point to the target grid cell; Δx is the spatial resolution of the grid cell in the horizontal direction; Δy is the spatial resolution of the grid cell in the vertical direction; i, j represent the position indices of the adjacent grid points in the spatial grid.

[0090] Spatial interpolation and gridding processing:

[0091] After unifying the spatial resolution, the climate factor data needs to be gridded through spatial interpolation methods to align spatially with the remote sensing image data and vegetation index data. Commonly used interpolation methods include Kriging interpolation and inverse distance weighted interpolation (IDW). In this embodiment, the inverse distance weighted interpolation (IDW) method is adopted, and its basic formula is:

[0092]

[0093] Where Z(s) is the interpolation result at the target position s, representing the predicted value of the climate factor at this position; X i is the climate factor value of the i-th known data point; d i is the spatial distance between the i-th known data point and the target position s; p is the distance weight exponent, usually a positive integer; n is the total number of known data points participating in the interpolation.

[0094] This method can generate climate factor grids that are spatially consistent with remote sensing image data by considering the influence of spatial distance.

[0095] Spatial division and regular grid data generation:

[0096] After completing spatial interpolation, the remote sensing image data, vegetation index data, and climate factor data are spatially divided using a preset grid size. For example, in this embodiment, all data are divided into grid cells of 30m x 30m. This unified spatial grid division ensures the consistency between different data layers. Specifically, the remote sensing image data and vegetation index data have been preprocessed at this resolution, while the climate factor data are converted into grid data with the same spatial resolution through spatial interpolation. This operation generates a regular grid data set, standardizing each data layer spatially.

[0097] Data normalization processing:

[0098] For each layer of data in the generated grid data set, a normalization method is used to convert the values to a unified numerical range. Through normalization, the dimensional differences between different data layers can be eliminated, enabling the comparison and analysis of each data layer on the same scale. The calculation formula for normalization is as follows:

[0099]

[0100] Where X min is the minimum value of the data set; X max is the maximum value of the data set; X is the value of the original data point; X norm is the result after normalization.

[0101] The result of the normalization process makes the values of each data layer fall within the range of [0, 1].

[0102] In some embodiments, if there is a large difference in the time scales between the remote sensing image data and the climate factor data (for example, the remote sensing image data is real-time monitoring data, while the climate factor data is annual average data), then it may be necessary to perform time interpolation on the climate factor data to match it with the remote sensing image data. In this case, linear interpolation or spline interpolation methods can be used, and the accuracy can be adjusted according to specific requirements.

[0103] In addition, in some embodiments, if there are many gaps in the remote sensing image data or the vegetation index, interpolation methods using neighboring pixels can be used to fill them, ensuring the integrity and consistency of the data set after spatial gridification.

[0104] Through the processing of this step, the remote sensing image data, vegetation index, and climate factor data can be unified in terms of spatial resolution, coordinate reference system, and numerical range, ensuring the comparability and consistency of the data. This processing effectively improves the collaborative utilization efficiency of multi-source remote sensing data and provides high-quality and standardized input data for subsequent regional ecological environment monitoring and analysis.

[0105] S4. Establish a non-linear regression model based on the climate factor data and vegetation index data;

[0106] For S4, the steps to establish a non-linear regression model based on the climate factor data and vegetation index data are as follows: First, based on the relationship between the climate factors and vegetation index data, combined with remote sensing image processing technology, regional ecological environment monitoring is carried out. In this process, through spatial grid and standardization processing of the climate factors and vegetation index, normalized data input is obtained. Next, a non-linear regression model is constructed to reveal the complex relationship between the climate factors and vegetation productivity.

[0107] Generally, when establishing a non-linear regression model, the climate factor data serves as the independent variable, while the vegetation index data serves as the dependent variable. Climate factors usually include environmental factors such as temperature, precipitation, and sunshine, and the vegetation index reflects the growth status of surface vegetation. In some embodiments, to enhance the prediction accuracy of the model, the climate factor data usually undergoes standardization processing before input, enabling data of different scales and dimensions to be unified into standardized values with the same dimension. This step is crucial for subsequent modeling and optimization processes.

[0108] In a possible implementation, the constructed non-linear regression model includes quadratic terms of climate factors and interaction terms between climate factors. Specifically, the expression of the model can be written as:

[0109] NPP = β0 + β1T + β2P + β3S + β4T 2 + β5P 2 + β6S 2 + β7T·P + β8P·S +

[0110] β9T·S + ∈;

[0111] Where, NPP is the net primary productivity of vegetation, calculated from the vegetation index data; T, P, and S respectively represent the standardized temperature, precipitation, and sunshine data; β0 is the constant term, β1 to β9 are the regression coefficients; ∈ is the error term. In the above equation, T 2 , P 2 and S 2 represent the quadratic terms of the climate factors, while T·P, P·S, and T·S represent the interaction terms between the climate factors.

[0112] In some embodiments, the parameters of the regression model are optimized by the least squares method. The least squares method is a commonly used mathematical optimization method, whose purpose is to determine the optimal regression coefficients by minimizing the sum of the squares of the errors between the predicted values and the actual observed values of the regression model. This optimization step can effectively improve the goodness of fit of the model and ensure the accuracy of the regression equation.

[0113] To further verify the effectiveness of the regression model, generally, the performance of the model is evaluated by goodness of fit (such as R 2 value, root mean square error, etc.). Through this verification process, the reliability and accuracy of the model in practical applications can be confirmed.

[0114] Specifically, the regression equation obtained by the above method can not only reflect the basic linear relationship between climate factors and vegetation productivity, but also capture the mutual influence and non - linear effects between climate factors, thus providing important technical support for the monitoring and evaluation of the regional ecological environment.

[0115] In one embodiment, the non - linear regression model can be further used for predicting ecological environment changes, especially for sensitivity analysis of changes in vegetation productivity, providing a scientific basis for fields such as precision agriculture and environmental protection.

[0116] In addition, as an option, in this regression model, the selection of climate factors and their processing methods can be adjusted according to actual needs to adapt to the requirements of ecological environment monitoring in different regions and different climate conditions.

[0117] S5. Extract topological feature data from the climate factor data based on the spatio - temporal topological data analysis method to identify the spatio - temporal change patterns of climate factors;

[0118] For S5, in the method for regional ecological environment monitoring based on remote sensing image processing, the identification of spatio - temporal change patterns of climate factor data is crucial. Since the spatio - temporal evolution process of climate factors usually has highly non - linear and complex spatial topological characteristics, it is necessary to adopt the spatio - temporal topological data analysis method to mine the stability patterns and spatial distribution characteristics of climate factors. Through topological analysis, the connectivity, circular structures and high - dimensional topological features in the climate factor data can be effectively extracted, providing reliable data support for subsequent environmental change monitoring and ecological prediction.

[0119] In this embodiment, the process of extracting topological feature data from climate factor data based on the spatio - temporal topological data analysis method includes the following steps:

[0120] Generally, first, it is necessary to perform structured processing on the climate factor data, and construct temperature, precipitation and sunshine duration into multi - dimensional spatio - temporal point cloud data. Specifically, assume that the climate factor data set is:

[0121] X = {(t i , x i , y i , z i ) | i = 1, 2, …, N};

[0122] Among them, X represents a multi-dimensional spatio-temporal point cloud data set composed of climate factors; i is a data index, representing the i-th climate factor observation point; N represents the total number of climate factor observation points; t i represents the time stamp of the i-th observation point, used to record the time position of the corresponding climate factor data; x i , y i respectively represent the longitude and latitude coordinates of the i-th observation point in the geographical coordinate system; z i represents the climate factor value of the i-th observation point, which can specifically be any one or more climate factor values such as temperature, precipitation, or sunshine hours.

[0123] In this way, the entire data set can be regarded as a high-dimensional point cloud, reflecting the change patterns of climate factors in time and space. In a possible implementation, for the above spatio-temporal point cloud data, the Persistent Homology algorithm is applied for topological analysis. Persistent Homology is a mathematical tool that can reveal the topological structure of data at different scales and is suitable for analyzing the local and global spatio-temporal distribution characteristics of climate factors. Specifically, a Vietoris-Rips complex VR ∈ (X) is constructed, where:

[0124]

[0125] Among them, VR ∈ (X) represents the Vietoris–Rips complex constructed based on the data set X under the condition of the threshold ∈; X is the point cloud data set on which the complex is constructed, usually a multi-dimensional spatio-temporal point set; σ is an arbitrary finite subset of X, called a simplex, where is the distance function between the point x i and x j ; ∈ is a distance threshold parameter used to control the scale size of complex construction.

[0126] Thus, a series of homology groups H k ;

[0127] Among them, H0 represents the connected components; H1 represents the circular structures, and H2 describes higher-dimensional topological features.

[0128] Specifically, when calculating topological features, the persistence algorithm is used to calculate the birth-death times of homology groups and generate a Persistence Barcode. Each bar (b i , d i ) in the barcode represents the existence time of a topological structure, where b i is the birth time and d i is the death time, and the length of the bar reflects the stability of the topological feature.

[0129] In some embodiments, to identify the stability patterns of climate factors, topological features with a duration exceeding a preset threshold τ can be extracted, i.e.:

[0130]

[0131] where τ can be set through statistical analysis or experience to ensure that the extracted topological features have statistical significance and physical meaning.

[0132] As an option, the topological stability index of climate factors, such as Topological Entropy, can be further calculated:

[0133] S = -∑ i p i log p i ;

[0134] where S represents the topological entropy value, which is used to measure the importance distribution and complexity of each topological feature in the entire topological structure; Σ i represents the summation over all topological features that meet the conditions; p i represents the normalized weight (or relative importance) of the i-th topological feature.

[0135] Topological entropy can measure the complexity of the spatio-temporal distribution of climate factors. A higher entropy value usually indicates that the climate factors change more violently, while a lower entropy value means that the climate factors are more stable in space and time.

[0136] In some embodiments, the topological feature data can be combined with other environmental variables (such as terrain, land use, etc.) to further analyze the driving forces of climate factors. For example, through the clustering analysis method, the topological feature data can be divided into different climate patterns, so as to identify regions with similar change trends and provide a scientific basis for ecosystem management.

[0137] Finally, the output topological feature data can be used to describe the spatio-temporal variation patterns of climate factors and serve as input variables for subsequent ecological environment assessment and prediction models. This method can not only identify the non-linear patterns of climate factors that are difficult to capture by traditional statistical methods but also reveal the potential interaction relationships between climate factors and ecosystems, providing technical support for precise environmental monitoring.

[0138] S6. Use a time series model to train the climate factor data and topological feature data, thereby making a long-term prediction of the change in regional vegetation productivity and outputting a prediction result.

[0139] For S6, to achieve the dynamic assessment of the regional ecological environment status and the prediction of change trends, after completing the previous processes such as spatial gridding and topological feature extraction of climate factor data, it is necessary to further adopt a modeling method with time series learning ability to achieve long-term prediction and analysis of regional vegetation productivity. The aforementioned topological analysis has fully extracted the high-order structural features of climate factors in the time and space dimensions, and spatial gridding ensures the consistency of data scale and format. On this basis, by using a time series model to fuse multi-dimensional climate data and its spatio-temporal topological features, the lagging effects and non-linear evolution relationships of environmental factors on vegetation productivity can be effectively captured.

[0140] In this embodiment, using a time series model to train the climate factor data and topological feature data for long-term prediction of the change in regional vegetation productivity specifically includes the following steps:

[0141] Generally, first, the spatially gridded and normalized climate factor data (including temperature, precipitation, sunshine hours, etc.) and their corresponding topological feature data (such as persistence, topological entropy, main loop structure index, etc.) need to be aligned according to a unified time step to construct a standardized time series input data set. As an option, when constructing the input, the input sample at each moment t can be defined as a vector:

[0142]

[0143] Where: X t represents the input vector at time t; represents the value of the i-th climate factor at time t, i = 1, 2,..., m; represents the value of the j-th topological feature at time t, j = 1, 2,..., n; m is the number of climate factor dimensions; n is the number of topological features.

[0144] In a possible implementation, a fixed-length time window w is defined as the input sequence span of the model. That is, for any prediction moment t, the data within the previous w time steps are combined as the model input, namely:

[0145] X t-w+1:t = [X t-w+1 , X t-w+2 , …, X t ;

[0146] Among them, X t-w+1:t represents the input sequence within a continuous time window with the current time being t and the length being w, which is the input relied on by the time series model for prediction at time t; X t represents the composite input vector corresponding to time step t; t is the current prediction reference time; w is the time window length, indicating the number of historical time steps referred to by the model during prediction; X t-w is the input data vector corresponding to the time point t - w.

[0147] Next, align the above-mentioned input sequence with the corresponding vegetation productivity indicators (such as NDVI, GPP, NPP, etc.) in time to construct a training sample set and a validation sample set. This training sample is defined as (X t-w+1:t , y t+k ), where y t+k represents the target vegetation productivity value at the prediction time point t + k, and k is the prediction step length.

[0148] Specifically, in this embodiment, the time series model adopted is a recurrent neural network architecture including long short-term memory units (LSTM, Long Short-Term Memory). LSTM can effectively model the long-term dependence problems in time series and is suitable for capturing the lag effect of climate factors on ecological indicators.

[0149] In a standard LSTM cell, the core calculations include an input gate, a forget gate, and an output gate, which are given by the following formulas respectively:

[0150] f t = σ(W f · [h t-1 , x t + b f );

[0151] i t = σ(W i · [h t-1 , x t + b i );

[0152]

[0153] o t = σ(W o · [h t-1 , x t + b o );

[0154] h t = o t *tanh(C t );

[0155] Where t represents the current time step; x t represents the input vector at time t; h t-1 represents the hidden state (also known as short-term memory) at the previous time step (i.e., t-1); C t-1 represents the cell state (also known as long-term memory) at the previous time step; f t is the forget gate vector; i t is the input gate vector; is the candidate memory value, which is the potential memory calculated from the input at the current time step; C t is the updated cell state, combining the previous memory and the current candidate; o t is the output gate vector, used to control the generation of the final output h t ; h t is the hidden state at the current time step and is also the output passed to the next time step; * represents element-wise multiplication; σ(·) represents the Sigmoid activation function; tanh(·) represents the hyperbolic tangent activation function; W f , W i , W C , W o are the weight matrices corresponding to the forget gate, input gate, candidate state, and output gate respectively; b f , b i , b C , b o are the bias terms for the above gates or states respectively.

[0156] During the model training process, the Root Mean Square Error (RMSE) is used as the loss function, and its definition is as follows:

[0157]

[0158] Where N represents the total number of samples, usually referring to the number of data points in the prediction set or validation set; y i represents the true value of the i-th sample; represents the model prediction value of the i-th sample.

[0159] The model parameters are continuously optimized through the error backpropagation algorithm until the loss function converges.

[0160] In some embodiments, to improve the generalization ability of the model, a regularization strategy (such as Dropout) can be introduced or multiple LSTM units can be integrated for multi-scale modeling; an attention mechanism (Attention) can also be introduced to enhance the model's ability to focus on important time steps, thereby improving the prediction accuracy.

[0161] After the model training is completed, the constructed model is applied to a new input data sequence, and the prediction results of the vegetation productivity in the future time period within the output area are output, further providing data support for the dynamic monitoring and sustainable management of the ecological environment.

[0162] S7. Conduct regional ecological environment analysis and evaluation based on the prediction results to provide a scientific basis for ecological environment monitoring;

[0163] For S7, after completing the structured processing of climate factor data extracted from remote sensing images, the construction and fusion of topological features, and the long-term prediction of regional vegetation productivity through a time series model, in order to achieve a scientific judgment of the regional ecological state and a responsive study of future change trends, it is necessary to further carry out a systematic evaluation and analysis of the ecological environment based on the prediction output. This analysis not only depends on the quantitative results output by the model, but also needs to combine the spatial heterogeneity and temporal evolution characteristics of the climate factors themselves, so as to form a set of comprehensive judgment bases for ecological monitoring, early warning, and policy support.

[0164] In this embodiment, regional ecological environment analysis and evaluation are carried out based on the prediction results to provide scientific support for ecological environment monitoring, specifically including the following technical steps:

[0165] Generally, first obtain the vegetation productivity prediction results output by the long short-term memory time series model (LSTM). This prediction result is the target variable value within the future k time steps predicted by the model based on the input time window X t-w+1:t and is denoted as:

[0166]

[0167] where represents the predicted value of the vegetation productivity at time t + k; k is the prediction step length, usually set according to the monitoring period; X t-w+1:t is the time window input sequence.

[0168] In a possible implementation, the model prediction value is compared with the historical observation value to calculate the relative change rate of the vegetation productivity. This change rate can be expressed as:

[0169]

[0170] where Δ tIndicates the change rate of vegetation productivity from time t to t + k; y t Indicates the actual observed value at time t; Is the predicted value.

[0171] As an option, to further evaluate the overall ecological evolution trend of the region, the Δ of different grid regions can be t Subjected to spatial statistical analysis to form a spatial distribution map for identifying sensitive areas of ecological degradation or restoration.

[0172] Specifically, in the present invention, an ecological risk level classification model is also constructed in combination with the spatio-temporal distribution characteristics of the climate factors themselves. This model comprehensively considers the following two types of input indicators:

[0173] The first type is the change rate of vegetation productivity Δ t , reflecting the direct result of the change in the ecological state;

[0174] The second type is spatio-temporal topological indicators, including topological entropy S = -∑p i logp i , persistence statistical mean, number of main topological classes (such as the number of H1 components), etc., for evaluating the complexity and fluctuation characteristics of the climate factors in the spatial structure.

[0175] In some embodiments, based on multi-index weighted clustering or fuzzy logic methods, the region is divided into multiple ecological risk level zones, such as high-risk zones, medium-risk zones, stable zones, etc. The risk level definition can be constructed with reference to the following rules:

[0176]

[0177] Among them, R(x, y) represents the ecological risk level corresponding to the spatial point (x, y); S is the topological entropy; Is the persistence mean; D clim Is the standard deviation or fluctuation intensity of the climate variable in this region; f(·) is a multi-factor decision function.

[0178] Subsequently, a multi-dimensional visualization analysis report including predicted trend charts, risk level charts, climate factor fluctuation charts, etc. is generated to enhance the perception and dissemination ability of the ecological evolution process.

[0179] In a possible technical implementation, this report is output in the form of a grid map with a consistent spatial resolution, and each grid contains: direction of change in vegetation productivity (positive / negative), change amplitude, variation level of climate factors, risk level, and indication of the evolution path, etc.

[0180] Finally, according to the above analysis results, a text of ecological protection suggestions is automatically generated, and its content includes but is not limited to:

[0181] Priority area ranking for vegetation restoration;

[0182] Recommendation of climate adaptation management measures (such as proposing water resource regulation strategies for high-temperature risk areas);

[0183] Early warning prompts for long-term evolution trends, etc.

[0184] Generally, this recommendation can generate structured output through a template mechanism or semantic enhancement generation in combination with an expert system, facilitating direct support for ecological policy formulation and regional intervention strategy deployment.

[0185] Please refer to the appendix Figure 2 , the present invention also provides a regional ecological environment monitoring system based on remote sensing image processing, including:

[0186] A data acquisition module, configured to acquire remote sensing image data of a target area and extract vegetation indices, as well as obtain climate factor data of the target area, where the climate factor data at least includes temperature, precipitation, and sunshine;

[0187] The data acquisition module is responsible for acquiring remote sensing image data of the target area, extracting vegetation indices from it, and obtaining important information reflecting the growth status of vegetation. In addition, this module also obtains climate factor data of the target area, which at least includes the main climate factors such as temperature, precipitation, and sunshine that affect vegetation growth. Through this module, the system can obtain complete spatial environment data, providing basic data for subsequent analysis and modeling.

[0188] A data processing module, configured to perform spatial gridding and standardization processing on the remote sensing image data, vegetation indices, and climate factor data to generate a raster data set with a unified spatial scale;

[0189] The data processing module performs spatial gridding and standardization processing on the acquired remote sensing image data, vegetation indices, and climate factor data to generate a raster data set with a unified spatial scale. In this way, data from different sources can be converted into a unified format, facilitating subsequent model training and analysis.

[0190] A non-linear regression modeling module, configured to construct a non-linear regression model based on the climate factor data and vegetation index data, and the model describes the relationship between vegetation productivity and climate factors through a regression equation including quadratic terms and interaction terms of climate factors;

[0191] The non-linear regression modeling module constructs a non-linear regression model through the analysis of climate factor data and vegetation index data. This model accurately describes the relationship between vegetation productivity and climate factors through a regression equation including quadratic terms and interaction terms of climate factors. This regression model can reveal the influence degree of different climate factors on vegetation productivity, providing necessary parameters for subsequent time series prediction and ecological assessment.

[0192] A spatio-temporal topology analysis module for extracting topological feature data from climate factor data and identifying its spatio-temporal change patterns;

[0193] The spatio-temporal topology analysis module is responsible for extracting topological feature data from climate factor data and identifying its spatio-temporal change patterns. By analyzing the spatio-temporal distribution characteristics of climate factors, this module helps the system identify the change laws of climate factors in the spatial and temporal dimensions, providing support for more accurate ecological assessment.

[0194] A time series prediction module for training climate factor data and topological feature data through a long short-term memory network to generate long-term prediction results of regional vegetation productivity;

[0195] After obtaining climate factor and topological feature data, the time series prediction module uses a long short-term memory network (LSTM) to train this data and generate long-term prediction results of regional vegetation productivity. By learning the non-linear relationship between climate factors and vegetation productivity through the model, the system can predict the change of vegetation productivity in the future for a period of time, and thus provide forward-looking information for regional ecological environment monitoring.

[0196] An ecological assessment module for analyzing the change trend of vegetation productivity based on the prediction results, dividing ecological risk regions, and outputting a visual report and management suggestions;

[0197] Based on the vegetation productivity prediction results output by the time series prediction module, the ecological assessment module analyzes its change trend, identifies regions of ecological degradation or restoration, and visually displays the analysis results. In addition, this module will also combine the impacts of climate factors and topological features to divide ecological risk level regions and output corresponding management suggestions according to the analysis results, such as the priority of vegetation restoration and climate adaptation management measures.

[0198] This system can achieve comprehensive monitoring and analysis of the regional ecological environment, with high accuracy and real-time performance, and can provide a scientific basis for ecological protection, resource management, and policy decision-making.

[0199] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for regional ecological environment monitoring based on remote sensing image processing, characterized in that, The following steps are involved: Collect remote sensing image data of the target area and extract vegetation index; Obtain climate factor data for the target area; Performing spatial gridding and standardization processing on the remote sensing image data, vegetation index and climate factor data; Establishing a nonlinear regression model based on the climate factor data and the vegetation index data; Extracting topological feature data from the climate factor data based on a spatiotemporal topological data analysis method to identify the spatiotemporal variation pattern of the climate factor; Using a time series model to train the climate factor data and topological feature data, thereby making long-term predictions of changes in regional vegetation productivity and outputting prediction results; Based on the prediction results, regional ecological environment analysis and assessment are carried out to provide a scientific basis for ecological environment monitoring.

2. The method for regional ecological environment monitoring based on remote sensing image processing according to claim 1, characterized in that The steps of collecting remote sensing image data of the target area and extracting vegetation index are as follows: Acquire remote sensing image data of the target area from ground or aerial photography equipment; Preprocessing the acquired remote sensing image data, wherein the preprocessing includes radiation correction, atmospheric correction and geometric correction; Select red light band and near infrared band data to calculate vegetation index, where the vegetation index is normalized vegetation index; The calculated vegetation index data are exported as standardized geographic information system raster data.

3. The regional ecological environment monitoring method based on remote sensing image processing according to claim 1, wherein The steps of obtaining the climate factor data of the target area are: Acquire climate factor data through ground meteorological stations, meteorological remote sensing data or historical meteorological databases, wherein the climate factor data at least includes temperature, precipitation and sunshine; Perform data cleaning on climate factor data, remove outliers and fill in missing data; The spatial interpolation method is used to convert the climate factor data into raster data format to match the remote sensing image data space, and the processed climate factor data is output.

4. The regional ecological environment monitoring method based on remote sensing image processing according to claim 1, characterized in that, The steps of spatially gridding and standardizing the remote sensing image data, vegetation index and climate factor data are as follows: Unify the spatial resolution and coordinate reference system of remote sensing image data, vegetation index and climate factor data; The spatial interpolation method is used to grid the climate factor data so that it can be spatially aligned with the remote sensing image data; The remote sensing image data, vegetation index and climate factor data are spatially divided according to the preset grid size to generate a regularized raster data set; Normalizing each data layer in the raster data set to convert the values of each data layer to the same value range; Output remote sensing image data, vegetation index and climate factor data that have been spatially gridded and standardized.

5. The regional ecological environment monitoring method based on remote sensing image processing according to claim 1, characterized in that The steps of establishing a nonlinear regression model based on the climate factor data and the vegetation index data are: The climate factor data after spatial gridding and standardization are used as independent variables; The vegetation index data after spatial gridding and standardization was used as the dependent variable; Constructing a nonlinear regression model, wherein the nonlinear regression model includes quadratic terms of climate factors and interaction terms between climate factors; Optimize the nonlinear regression model parameters through the least squares method and determine the regression coefficients; Verify the goodness of fit of the nonlinear regression model and output the regression equation used to describe the relationship between climate factors and vegetation productivity.

6. The regional ecological environment monitoring method based on remote sensing image processing according to claim 5, characterized in that The nonlinear regression equation is expressed as: NPP = β0 + β1T + β2P + β3S + β4T 2 + β5P 2 + β6S 2 + β7T·P + β8P·S + β9T·S+∈; Among them, NPP is the net primary productivity of vegetation, which is calculated from vegetation index data; T, P, and S respectively represent the standardized temperature, precipitation, and sunshine data; β0 is the constant term, and β1 to β9 are regression coefficients; ∈ is the error term.

7. The regional ecological environment monitoring method based on remote sensing image processing according to claim 1, characterized in that, The steps of extracting topological feature data from the climate factor data based on the spatio-temporal topological data analysis method are as follows: Construct temperature, precipitation, and sunshine hours in the climate factor data into multi-dimensional spatio-temporal point cloud data; Apply the persistent homology algorithm to perform topological analysis on the spatio-temporal point cloud data and calculate its persistent homology group; Generate a persistence bar chart according to the birth-death time of the homology group to identify the connectivity and circular structure in the spatio-temporal distribution of climate factors; Extract the topological features in the bar chart whose persistence threshold exceeds the preset value as the stability pattern of climate factors; Output the topological feature data for describing the spatio-temporal change pattern of climate factors.

8. The regional ecological environment monitoring method based on remote sensing image processing according to claim 1, characterized in that, The steps of training the climate factor data and topological feature data using a time series model are as follows: Align the climate factor data and topological feature data after spatial gridification and standardization in time series to construct a multi-dimensional input data set; Define the input window length of the time series model and divide the input data set into a training set and a validation set; Construct a time series model containing long short-term memory units, and the time series model recursively processes input data through time steps; Adopt the root mean square error as the loss function and optimize the parameters of the time series model through the backpropagation algorithm; Verify the prediction accuracy of the time series model and output the trained time series model for long-term prediction of regional vegetation productivity.

9. The regional ecological environment monitoring method based on remote sensing image processing according to claim 1, characterized in that The steps of providing a scientific basis for ecological environment monitoring based on the regional ecological environment analysis and evaluation of the prediction results are as follows: Obtain the regional vegetation productivity prediction results output by the time series model; Calculate the vegetation productivity change rate based on the prediction results and historical data, and analyze the regional ecological degradation or restoration trend; Combine the spatio-temporal distribution characteristics of climate factors to divide the ecological risk level area; Generate a visual analysis report containing the vegetation productivity change trend and ecological risk level; Output ecological protection suggestions according to the analysis results, and the ecological protection suggestions include the priority of vegetation restoration and climate adaptation management measures.

10. A regional ecological environment monitoring system based on remote sensing image processing, which is applied to the regional ecological environment monitoring method based on remote sensing image processing according to any one of claims 1-9, and is characterized in that, Including: A data acquisition module for collecting remote sensing image data of the target area and extracting vegetation indices, and obtaining climate factor data of the target area, where the climate factor data at least includes temperature, precipitation, and sunshine; A data processing module for performing spatial gridification and standardization processing on the remote sensing image data, vegetation indices, and climate factor data to generate a raster data set with a unified spatial scale; A non-linear regression modeling module for constructing a non-linear regression model based on the climate factor data and vegetation index data, and the model describes the relationship between vegetation productivity and climate factors through a regression equation including quadratic terms and interaction terms of climate factors; A spatio-temporal topological analysis module for extracting topological feature data from climate factor data and identifying its spatio-temporal change pattern; A time series prediction module for training climate factor data and topological feature data through a long short-term memory network to generate long-term prediction results of regional vegetation productivity; Ecological assessment module, which is used to analyze the changing trend of vegetation productivity based on the prediction results, divide ecological risk areas, and output a visualization report and management suggestions.

Images