Lake and reservoir chlorophyll concentration temporal and spatial change analysis method based on multi-source remote sensing

Abstract

The invention relates to a lake and reservoir chlorophyll concentration temporal and spatial change analysis method based on multi-source remote sensing. The method comprises the following steps: cooperatively collecting multispectral images and real-time water sample data through remote sensing and ground equipment, and obtaining a chlorophyll concentration measured value with geographic coordinates and timestamps; training a convolutional neural network inversion model by taking the measured value as a label, and processing the image to obtain a chlorophyll concentration distribution matrix; decomposing dynamic components through spectral analysis, extracting hydrological related subset data and optimizing to obtain a heterogeneity feature vector, and filtering interference through machine learning to obtain a pure signal sequence if sediment suspension disturbance exceeds the standard; calculating environment and climate influence weights based on the sequence, correcting the concentration, and generating optimized space-time distribution representation after validity score verification; finally, key monitoring points are extracted through differential monitoring grid division, and a chlorophyll concentration monitoring parameter set is formed. According to the method, the time-space continuity and accuracy of the monitored chlorophyll concentration data are guaranteed, and the monitoring efficiency is improved.

Description

Technical Field

[0001] This invention belongs to the field of water quality remote sensing monitoring technology, and in particular relates to a method for analyzing the spatiotemporal changes of chlorophyll concentration in lakes and reservoirs based on multi-source remote sensing. Background Technology

[0002] As important carriers of freshwater resources, the water quality of lakes and reservoirs directly affects the water ecological balance, drinking water safety, and the sustainable economic and social development of the region. Chlorophyll concentration, as a core indicator reflecting the degree of eutrophication of water bodies, is crucial for water quality monitoring and ecological management by accurately capturing its spatiotemporal dynamic changes. Traditional chlorophyll concentration monitoring relies heavily on ground-based fixed-point sampling and detection. Although it can obtain high-precision data at a single point, it has limitations such as limited coverage, time and labor consumption, and difficulty in reflecting the spatiotemporal heterogeneity of water bodies. While single remote sensing technology has achieved large-scale and rapid monitoring, it is affected by factors such as suspended sediment, cloud shadow interference, and spectral mixing effects, resulting in insufficient accuracy of inversion models. Furthermore, existing technologies have shortcomings in dynamic component separation, concentration correction in high-interference areas, construction of spatiotemporally continuous data, and optimized layout of monitoring points, making it difficult to meet the actual needs of high-precision and comprehensive spatiotemporal variation analysis of chlorophyll concentration in lakes and reservoirs. Summary of the Invention

[0003] Therefore, it is necessary to provide a method for analyzing the spatiotemporal changes of chlorophyll concentration in lakes and reservoirs based on multi-source remote sensing, which can improve the spatiotemporal continuity, accuracy, and monitoring efficiency of chlorophyll concentration data monitoring, in order to address the above-mentioned technical problems.

[0004] Firstly, this application provides a method for analyzing the spatiotemporal variation of chlorophyll concentration in lakes and reservoirs based on multi-source remote sensing, including:

[0005] Multispectral images and real-time water sample data of lakes and reservoirs were collected. Chlorophyll concentration of the real-time water sample data was detected to obtain the measured values ​​of chlorophyll concentration of the water sample with geographic coordinates and timestamps.

[0006] A convolutional neural network inversion model is trained based on the multi-dimensional spectral feature vectors corresponding to geographic coordinates and the measured values ​​of chlorophyll concentration in water samples. Multi-spectral images are then input into the trained inversion model to obtain the chlorophyll concentration distribution matrix.

[0007] The chlorophyll concentration distribution matrix was decomposed into dynamic components corresponding to seasonal and diurnal variations using spectral analysis. Subset data related to hydrological characteristics were extracted from the decomposed dynamic components, and the subset data were adjusted and optimized to obtain heterogeneous feature vectors.

[0008] If the sediment suspension disturbance component in the heterogeneous feature vector exceeds a preset threshold, then the subset data is filtered for interference signals using a machine learning classification method to obtain a pure chlorophyll concentration signal sequence.

[0009] The influence weights of environment and climate are calculated based on the pure chlorophyll concentration signal sequence, and the chlorophyll concentration sequence is corrected. The effectiveness score is calculated based on the corrected chlorophyll concentration sequence. If the effectiveness score reaches the preset threshold, the optimized spatiotemporal distribution representation of chlorophyll concentration is generated.

[0010] The optimized spatiotemporal distribution of chlorophyll concentration was represented by dividing the spatiotemporal distribution data into a differentiated monitoring grid and extracting the location information of key monitoring points to obtain a set of chlorophyll concentration monitoring parameters.

[0011] In one embodiment, a convolutional neural network inversion model is trained based on the multi-dimensional spectral feature vector corresponding to geographic coordinates and the measured chlorophyll concentration of water samples. Multi-spectral images are then input into the trained inversion model to obtain a chlorophyll concentration distribution matrix, including:

[0012] Based on the spatiotemporal heterogeneity of lakes and reservoirs, pixel-level spectral information is extracted from multi-spectral images to construct multi-dimensional spectral feature vectors that correspond one-to-one with geographic coordinates.

[0013] Using multi-dimensional spectral feature vectors as input features and measured values ​​of chlorophyll concentration in water samples as labels, a pre-defined convolutional neural network model is trained to obtain a converged chlorophyll concentration inversion model.

[0014] The full-pixel spectral information of multi-band images is input into the chlorophyll concentration inversion model, and the model inference outputs a preliminary estimate of the chlorophyll concentration corresponding to the geographic coordinates of each pixel and the image timestamp.

[0015] A spatiotemporal data fusion algorithm was used to interpolate and complete the preliminary estimates of chlorophyll concentration corresponding to adjacent timestamps, resulting in a spatiotemporally continuous chlorophyll concentration distribution matrix covering the entire water area.

[0016] In one embodiment, the chlorophyll concentration distribution matrix is ​​decomposed into dynamic components corresponding to seasonal and diurnal variations using spectral analysis. Subset data related to hydrological characteristics are extracted from the decomposed dynamic components, and the subset data is adjusted and optimized to obtain a heterogeneous feature vector, including:

[0017] The chlorophyll concentration distribution matrix was dynamically decomposed using spectral analysis to separate seasonal variation data reflecting long-term patterns and diurnal variation data reflecting short-term fluctuations.

[0018] Based on preset hydrological characteristic indicators, corresponding subset data are extracted from seasonal variation data and diurnal variation data to obtain an initial feature vector.

[0019] Based on the initial feature vector, a heterogeneous feature vector containing spatial heterogeneity, temporal heterogeneity, and hydrological correlation dimensions is constructed. The distribution characteristics data of the heterogeneous feature vector are obtained through statistical analysis methods. The distribution characteristics data include the distribution mean, variance, extreme values, and distribution density.

[0020] Using distribution characteristic data as input, and combining preset association rules for hydrological and heterogeneous characteristics, a dynamic adjustment model that is dynamically associated with hydrological characteristics is generated.

[0021] The heterogeneous feature vectors are input into the dynamically adjusted model for mapping processing to obtain the feature mapping results.

[0022] Determine whether the feature mapping result is within the preset threshold range. If not, iterate and optimize the parameters of the dynamically adjusted model based on the deviation between the feature mapping result and the threshold range to determine the final heterogeneous feature vector.

[0023] In one embodiment, the influence weights of environment and climate are calculated based on a pure chlorophyll concentration signal sequence, and the chlorophyll concentration sequence is corrected. An effectiveness score is calculated based on the corrected chlorophyll concentration sequence. If the effectiveness score reaches a preset threshold, an optimized spatiotemporal distribution representation of chlorophyll concentration is generated, including:

[0024] Based on the pure chlorophyll concentration signal sequence combined with the carried geographic coordinates and timestamp information, the corresponding geographic environment difference factors and climate condition change factors are extracted.

[0025] Correlation analysis was used to determine the correlation between geographical environment difference factors, climate condition change factors and pure chlorophyll concentration signals. Based on the correlation, a weighted fusion algorithm was used to calculate the influence weights of geographical environment difference and climate condition change, and the weight set was obtained by integration.

[0026] Based on the distribution characteristics of each weight in the weight set and combined with the spatial fluctuation pattern of the pure chlorophyll concentration signal sequence, a spatial clustering algorithm is used to extract the local interference pattern of the high interference area. The geographical coordinate range of the high interference area is determined by pattern matching, and the high interference area is identified.

[0027] For regions with high interference, the corresponding regions are identified, and the concentration correction value of the corresponding regions is calculated by combining the influence coefficients of each weight in the weight set on chlorophyll concentration using a weighted correction formula. For regions without high interference, the pure chlorophyll concentration signal is directly used to obtain the corrected chlorophyll concentration sequence.

[0028] The values ​​of each spatiotemporal node in the corrected chlorophyll concentration sequence are compared with the real-time water sample data with the corresponding geographic coordinates and timestamps. The comparison difference values ​​of each spatiotemporal node are obtained. Based on the comparison difference values, a comparison difference distribution covering the entire water area is constructed. The comparison difference distribution includes a spatial distribution map and a temporal variation curve.

[0029] Based on the comparison of the difference distribution, the mean, standard deviation, maximum deviation and the percentage of qualified nodes are calculated. A comprehensive scoring model is used to obtain the effectiveness score of the interference separation strategy. Qualified nodes are those whose difference values ​​are within the preset allowable error range.

[0030] If the effectiveness score is lower than the preset effectiveness score threshold, the weights in the weight set are iteratively adjusted based on the weight contribution of the high deviation area in the comparison difference distribution until the effectiveness score reaches the effectiveness score threshold, thus obtaining the optimized spatiotemporal distribution representation of chlorophyll concentration; the effectiveness score threshold is calibrated based on historical verification data and water quality monitoring accuracy requirements.

[0031] In one embodiment, the weights of the impact of geographical environmental differences and the impact of climate condition changes are calculated using the following formula:

[0032]

[0033]

[0034] in, This indicates the weighting of the impact of geographical environmental differences. This indicates the weight of the impact of climate change. This represents the weighting balance coefficient. Pre-marked according to the type of lake or reservoir. The average Pearson correlation coefficient represents the set of geographical environmental difference factors and the pure chlorophyll concentration signal. , This indicates the number of factors contributing to geographical environmental differences. Indicates the first The correlation coefficients between geographical environmental factors and pure chlorophyll concentration signals The average Pearson correlation coefficient between the set of climate condition change factors and the pure chlorophyll concentration signal. , This indicates the number of factors affecting climate change. Indicates the first The correlation coefficients between the climate factors and the pure chlorophyll concentration signal The comprehensive information entropy represents the set of factors contributing to geographical environmental differences. The comprehensive information entropy represents the set of factors affecting climate change.

[0035] In one embodiment, the concentration correction value is calculated using the following formula:

[0036]

[0037] in, This indicates the concentration correction value. This represents the raw value of the pure chlorophyll concentration signal within the high-interference region. Indicates the geographical environment adaptability coefficient. , This represents the average slope of the target area. This represents the average slope of the entire body of water. , Indicates the climate adaptability coefficient. , The average temperature of the target area over the past 30 days indicates... This indicates the average temperature of the entire water body over the past 30 days. , Indicates the interference intensity coefficient in the high interference region. , The standard deviation of the local interference pattern in the target region is represented by the following: The standard deviation of the pure chlorophyll concentration signal across the entire water body is represented. , This indicates the weighting of the impact of geographical environmental differences. This indicates the weight of the impact of climate change.

[0038] In one embodiment, the optimized spatiotemporal distribution representation of chlorophyll concentration is divided into spatiotemporal distribution data using a differentiated monitoring grid, and the location information of key monitoring points is extracted to obtain a set of chlorophyll concentration monitoring parameters, including:

[0039] The optimized spatiotemporal distribution of chlorophyll concentration was represented by a differentiated monitoring grid division method, generating a differentiated monitoring grid covering the entire water area and corresponding spatiotemporal distribution data of chlorophyll concentration for each grid unit.

[0040] Based on differentiated monitoring grids and spatiotemporal distribution data of chlorophyll concentration, combined with the coefficient of variation of chlorophyll concentration within the grid unit, spatial coverage requirements, and monitoring cost constraints, the geographic coordinate information of key monitoring points is extracted to form a preliminary monitoring point dataset containing the location of the monitoring point, the grid unit to which it belongs, and the initial monitoring period.

[0041] The preliminary monitoring point dataset is integrated with the mean chlorophyll concentration, concentration change rate, and concentration extreme value corresponding to each monitoring point to construct a dynamic monitoring parameter matrix.

[0042] Principal component analysis was used to remove redundant parameters from the dynamic monitoring parameter matrix, and parameters with information contribution higher than a preset threshold were retained to determine the core monitoring parameter set for chlorophyll concentration.

[0043] Based on the concentration sensitivity of each monitoring point in the core monitoring parameter set, grid densification rules are set for areas with high information gain of core monitoring parameters to generate extended monitoring grids.

[0044] By interpolating the optimized spatiotemporal distribution representation of chlorophyll concentration using an extended monitoring grid, refined concentration distribution feature data including concentration gradient, spatial clustering features, and temporal abrupt change nodes are obtained.

[0045] Based on refined concentration distribution characteristic data, the correlation coefficient method and spatiotemporal autocorrelation analysis were used to calculate the concentration correlation strength between key monitoring points at different time scales, and the correlation strength matrix between monitoring points was obtained.

[0046] If the correlation strength between monitoring points in the correlation strength matrix is ​​lower than the preset threshold, then the resolution, boundary division, and densification rules of the differentiated monitoring grid are adjusted based on the high heterogeneity distribution in the refined concentration distribution feature data to generate a chlorophyll concentration monitoring parameter set.

[0047] Secondly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0048] Thirdly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method.

[0049] The aforementioned method, computer equipment, and storage medium for analyzing the spatiotemporal variation of chlorophyll concentration in lakes and reservoirs based on multi-source remote sensing first involves collaboratively acquiring multispectral images and real-time water sample data of lake and reservoir water areas using remote sensing equipment and ground monitoring equipment. Chlorophyll concentration is then measured on the real-time water sample data to obtain measured values ​​of chlorophyll concentration carrying geographic coordinates and timestamps. Subsequently, based on the multi-dimensional spectral feature vectors corresponding to the geographic coordinates, a convolutional neural network inversion model is trained using the measured chlorophyll concentration values ​​of the water samples as labels. The multispectral images are input into the trained inversion model to generate a chlorophyll concentration distribution matrix. Next, spectral analysis is applied to this concentration distribution matrix to decompose the dynamic components corresponding to seasonal and diurnal variations. Subset data related to hydrological characteristics are then extracted from the decomposed dynamic components. After adjustment and optimization, a heterogeneous feature vector is obtained. Then, it is determined whether the sediment suspension disturbance component in the heterogeneous feature vector exceeds a preset threshold. If so, the interference signal is filtered through the above subset data using a machine learning classification method to obtain a pure chlorophyll concentration signal sequence. Based on this pure chlorophyll concentration signal sequence, the influence weights of the environment and climate are calculated, and the chlorophyll concentration sequence is corrected using the weights. Then, the effectiveness score of the interference separation strategy is calculated based on the corrected chlorophyll concentration sequence. When the score reaches a preset threshold, an optimized spatiotemporal distribution representation of chlorophyll concentration is generated. Finally, the spatiotemporal distribution data of the optimized spatiotemporal distribution representation of chlorophyll concentration is processed using a differentiated monitoring grid division method to extract the location information of key monitoring points, and finally, a chlorophyll concentration monitoring parameter set is formed. This method realizes a closed-loop processing of chlorophyll concentration data from acquisition, modeling, denoising, correction to monitoring layout optimization. It effectively solves the problems of limited coverage of traditional monitoring, insufficient accuracy of single remote sensing inversion, and poor adaptability to spatiotemporal heterogeneity. It not only ensures the spatiotemporal continuity and accuracy of chlorophyll concentration data, but also improves monitoring efficiency through optimized monitoring point layout. It provides reliable technical support for high-precision spatiotemporal variation analysis of chlorophyll concentration in lakes and reservoirs and water ecological management. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 A flowchart illustrating the spatiotemporal variation analysis method for chlorophyll concentration in lakes and reservoirs based on multi-source remote sensing, provided in an embodiment of the present invention.

[0052] Figure 2The present invention provides a flowchart for calculating the influence weights of environment and climate based on a pure chlorophyll concentration signal sequence and correcting the chlorophyll concentration sequence. The effectiveness score is calculated based on the corrected chlorophyll concentration sequence. If the effectiveness score reaches a preset threshold, an optimized spatiotemporal distribution representation of chlorophyll concentration is generated. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0054] In one embodiment, such as Figure 1 As shown, this application provides a method for analyzing the spatiotemporal variation of chlorophyll concentration in lakes and reservoirs based on multi-source remote sensing, which may include the following steps:

[0055] Step S101: Collect multispectral images and real-time water sample data of lakes and reservoirs, and detect chlorophyll concentration in the real-time water sample data to obtain the measured chlorophyll concentration of the water sample carrying geographic coordinates and timestamps.

[0056] Specifically, multispectral image data of the entire lake and reservoir area is collected using remote sensing equipment (such as multispectral satellites and UAV remote sensing payloads). Simultaneously, real-time water sample data is collected at sampling points deployed spatially and uniformly within the water area using ground monitoring equipment. The number and distribution of sampling points must match the spatial resolution and coverage of the multispectral imagery. The collected real-time water sample data are then quantitatively analyzed for chlorophyll concentration using standard laboratory testing methods (such as spectrophotometry and high-performance liquid chromatography). The test results for each water sample are linked to the geographical coordinates (latitude and longitude) and sampling timestamp of the corresponding sampling point, ultimately forming a measured chlorophyll concentration value for each water sample carrying the geographical coordinates, timestamp, and chlorophyll concentration value.

[0057] Step S102: Train a convolutional neural network inversion model based on the multi-dimensional spectral feature vector corresponding to the geographic coordinates and the measured value of chlorophyll concentration in the water sample. Input the multi-spectral image into the trained inversion model to obtain the chlorophyll concentration distribution matrix.

[0058] Pixel-level spectral information is extracted from the acquired multispectral images, including reflectance, band ratio, and texture features of each pixel in different spectral bands. Combined with the geographic coordinates of each pixel, a multi-dimensional spectral feature vector corresponding to each geographic coordinate is constructed. This multi-dimensional spectral feature vector is used as the model input, and the measured chlorophyll concentration of the water sample is used as the output label. A pre-set convolutional neural network model is iteratively trained. By adjusting the model's convolutional kernel size, pooling window size, activation function type, and number of iterations, the model's loss function value is reduced to a preset convergence threshold, resulting in a convergent and stable chlorophyll concentration inversion model. The complete full-pixel spectral information of the multispectral images is input into the trained inversion model. Through forward inference, the model outputs an estimated chlorophyll concentration value for each pixel. Combined with the pixel's geographic coordinates and image timestamps, a chlorophyll concentration distribution matrix covering the entire water body is formed.

[0059] Step S103: The chlorophyll concentration distribution matrix is ​​decomposed into dynamic components corresponding to seasonal and diurnal variations using spectral analysis. Subset data related to hydrological characteristics are extracted from the decomposed dynamic components, and the subset data is adjusted and optimized to obtain a heterogeneous feature vector.

[0060] Spectral analysis methods (such as Fourier spectral analysis and wavelet analysis) were used to dynamically decompose the obtained chlorophyll concentration distribution matrix. Based on the time-scale characteristics corresponding to different frequency components, seasonal variation data reflecting long-term variation patterns and diurnal variation data reflecting short-term fluctuation characteristics were separated. Based on preset hydrological characteristic indicators (including water level, flow velocity, water depth, turbidity, etc.), a subset of data strongly correlated with each hydrological indicator was selected from the seasonal and diurnal variation data through correlation analysis. Based on this subset of data, an initial feature vector containing spatial heterogeneity, temporal heterogeneity, and hydrological correlation attributes was constructed. Through statistical analysis, the distribution characteristics data such as the distribution mean, variance, extreme values, and distribution density of the initial feature vector were obtained. Combined with preset hydrological feature and heterogeneous feature correlation rules, a dynamic adjustment model was constructed to map and optimize the initial feature vector, finally obtaining the heterogeneous feature vector after dynamic component adjustment.

[0061] Step S104: Determine whether the sediment suspension disturbance component in the heterogeneous feature vector exceeds a preset threshold. If so, filter the interference signal of the subset data using a machine learning classification method to obtain a pure chlorophyll concentration signal sequence.

[0062] The sediment suspension disturbance component is extracted from the obtained heterogeneous feature vector. This component is calculated using the dimension data related to the sediment reflectance spectrum in the feature vector. The extracted sediment suspension disturbance component is compared with a preset threshold (calibrated based on historical sediment content monitoring data of the water area and the accuracy requirements of chlorophyll concentration inversion) to determine whether there is significant sediment interference. If the sediment suspension disturbance component exceeds the preset threshold, the extracted hydrologically relevant subset data is standardized and preprocessed to eliminate dimensional differences. Subsequently, a machine learning classification model (such as random forest or support vector machine) is constructed and trained with "sediment interference signal" and "pure chlorophyll signal" as classification targets. The preprocessed subset data is input into the trained model, and the sediment suspension interference signal is separated through classification. After removing the interference signal, the remaining data is smoothed to obtain a pure chlorophyll concentration signal sequence without sediment interference.

[0063] Step S105: Calculate the influence weights of environment and climate based on the pure chlorophyll concentration signal sequence and correct the chlorophyll concentration sequence. Calculate the effectiveness score based on the corrected chlorophyll concentration sequence. If the effectiveness score reaches a preset threshold, generate an optimized spatiotemporal distribution representation of chlorophyll concentration.

[0064] Based on the geographic coordinates and timestamps carried by the obtained pure chlorophyll concentration signal sequence, corresponding geographic environmental difference factors (such as topographic slope, shoreline type, and surrounding land use type) and climate condition change factors (such as temperature, precipitation, sunshine duration, and wind speed) are extracted. Correlation analysis is used to determine the degree of association between each factor and the pure chlorophyll concentration signal. Weighted fusion algorithms (such as the analytic hierarchy process and entropy weighting) are used to calculate the influence weight of each factor, integrating them into a weight set of geographic environmental difference influence weight and climate condition change influence weight. For regions corresponding to high-weight factors in the weight set, local patterns in high-interference regions are identified and extracted. A weighted correction formula is used to calculate the concentration correction value for these regions. Non-high-interference regions directly use the pure chlorophyll concentration signal, resulting in a corrected chlorophyll concentration sequence. The corrected chlorophyll concentration sequence is then compared point-by-point with real-time water sample data. Based on the difference values, the mean, standard deviation, maximum deviation, and percentage of qualified nodes are calculated. A comprehensive scoring model is used to obtain an effectiveness score for the interference separation strategy. A preset validity score threshold (calibrated based on historical validation data and water quality monitoring accuracy requirements) is set. If the score reaches the threshold, an optimized spatiotemporal distribution representation of chlorophyll concentration with spatiotemporal continuity and meeting accuracy standards is generated.

[0065] Step S106: The optimized spatiotemporal distribution representation of chlorophyll concentration is divided into spatiotemporal distribution data using a differentiated monitoring grid and the location information of key monitoring points is extracted to obtain a set of chlorophyll concentration monitoring parameters.

[0066] A differentiated monitoring grid method was used to represent the optimized spatiotemporal distribution of chlorophyll concentration. The grid resolution was set according to the spatial heterogeneity of concentration, with higher resolution for high-heterogeneity areas than for low-heterogeneity areas. This generated a differentiated monitoring grid covering the entire water area and corresponding spatiotemporal distribution data of chlorophyll concentration for each grid unit. Based on this differentiated monitoring grid and the chlorophyll concentration data of each grid unit, combined with the coefficient of variation of chlorophyll concentration within the grid unit, spatial coverage requirements, and monitoring cost constraints, the geographic coordinate information of key monitoring points was extracted to form a preliminary monitoring point dataset containing the location of each monitoring point, its grid unit, and the initial monitoring period. The mean chlorophyll concentration, rate of change of concentration, and extreme concentration values ​​of each monitoring point in the preliminary monitoring point dataset were integrated to construct a dynamic monitoring parameter matrix with dimensions of "monitoring point - monitoring indicator - time series". Principal component analysis was used to remove redundant parameters with information contribution below a preset threshold, thus determining the core monitoring parameter set. Based on the core monitoring parameter set and the concentration sensitivity of each monitoring point, grid densification rules are set in areas with high information gain of core parameters to generate an extended monitoring grid. This grid is then used to interpolate and refine the optimized spatiotemporal distribution data, resulting in refined concentration distribution characteristic data. The Pearson correlation coefficient method and spatiotemporal autocorrelation analysis (Moran's I index) are used to calculate the concentration correlation strength between each monitoring point. If there are monitoring points with correlation strength below a preset threshold, the grid resolution, boundary division, and densification rules are adjusted to re-optimize the monitoring point layout, ultimately forming a chlorophyll concentration monitoring parameter set that includes monitoring point location, core monitoring parameters, and monitoring frequency.

[0067] The aforementioned method for analyzing the spatiotemporal variation of chlorophyll concentration in lakes and reservoirs based on multi-source remote sensing involves collaboratively acquiring multi-spectral images and real-time water sample data using remote sensing and ground-based equipment to obtain measured chlorophyll concentration values ​​with geographic coordinates and timestamps. A convolutional neural network inversion model is trained using these measured values ​​as labels, and the images are processed to obtain a chlorophyll concentration distribution matrix. Dynamic components are decomposed through spectral analysis, hydrologically relevant subset data are extracted, and heterogeneous feature vectors are obtained through optimization. If sediment suspension disturbance exceeds the standard, interference is filtered through machine learning to obtain a clean signal sequence. Based on this sequence, the weights of environmental and climate influences are calculated, and the concentration is corrected. After validity scoring verification, an optimized spatiotemporal distribution representation is generated. Finally, key monitoring points are extracted through differentiated monitoring grid division to form a chlorophyll concentration monitoring parameter set. This method achieves closed-loop processing from data acquisition, modeling, denoising, correction to monitoring layout optimization, solving problems such as limited monitoring coverage and insufficient accuracy of single remote sensing inversion in traditional methods. It ensures the spatiotemporal continuity and accuracy of data, improves monitoring efficiency, and provides reliable support for lake and reservoir water quality monitoring and water ecological management.

[0068] In one embodiment, a convolutional neural network inversion model is trained based on the multi-dimensional spectral feature vector corresponding to geographic coordinates and the measured chlorophyll concentration of water samples. Multi-spectral images are then input into the trained inversion model to obtain a chlorophyll concentration distribution matrix. This process may include the following steps:

[0069] Step S201: Based on the spatiotemporal heterogeneity of lake and reservoir water areas, pixel-level spectral information is extracted from multi-spectral images to construct a multi-dimensional spectral feature vector that corresponds one-to-one with geographic coordinates.

[0070] Step S202: Using multi-dimensional spectral feature vectors as input features and measured values ​​of chlorophyll concentration in water samples as labels, a pre-defined convolutional neural network model is trained to obtain a converged chlorophyll concentration inversion model.

[0071] Step S203: Input the full-pixel spectral information of the multi-spectral image into the chlorophyll concentration inversion model, and output the preliminary estimate of chlorophyll concentration corresponding to the geographic coordinates and image timestamps of each pixel through model inference.

[0072] Step S204: The spatiotemporal data fusion algorithm is used to interpolate and complete the preliminary estimates of chlorophyll concentration corresponding to adjacent timestamps to obtain a spatiotemporally continuous chlorophyll concentration distribution matrix covering the entire water area.

[0073] Specifically, based on the spatiotemporal heterogeneity of lakes and reservoirs, pixel-level spectral information is extracted from collected multispectral images. The extracted information includes reflectance, band ratios, and texture features of each pixel in different spectral bands. Based on this, a multi-dimensional spectral feature vector corresponding to each geographic coordinate is constructed. Using this multi-dimensional spectral feature vector as input, and simultaneously acquired measured chlorophyll concentration values ​​of water samples carrying geographic coordinates and timestamps as labels, a pre-set convolutional neural network model is iteratively trained. By adjusting the model parameters to reduce the loss function value to a pre-set convergence threshold, a convergent and stable chlorophyll concentration inversion model is obtained. The full-pixel spectral information of the multispectral images is input into the trained chlorophyll concentration inversion model. Through forward inference, the model outputs preliminary estimates of chlorophyll concentration corresponding to each pixel's geographic coordinates and image timestamps. Then, a spatiotemporal data fusion algorithm is used to interpolate and complete the preliminary estimates of chlorophyll concentration corresponding to adjacent timestamps, filling in missing data areas. Finally, a spatiotemporally continuous chlorophyll concentration distribution matrix covering the entire water area is obtained.

[0074] This embodiment incorporates the spatiotemporal heterogeneity of water bodies into the feature construction stage, enabling a precise correspondence between multi-dimensional spectral feature vectors and geographic coordinates, thus enhancing the correlation between input features and chlorophyll concentration. Using measured values ​​as labels to train the convolutional neural network inversion model ensures the accuracy of the model inversion. By inputting full-pixel spectral information into the model, a preliminary estimate of chlorophyll concentration across the entire water body is achieved. Combined with interpolation completion processing using spatiotemporal data fusion algorithms, the potential data breakpoint problem in single remote sensing images is effectively resolved. The resulting chlorophyll concentration distribution matrix possesses both full water body coverage and spatiotemporal continuity.

[0075] In one embodiment, the chlorophyll concentration distribution matrix is ​​decomposed into dynamic components corresponding to seasonal and diurnal variations using spectral analysis. Subset data related to hydrological characteristics is extracted from the decomposed dynamic components, and the subset data is adjusted and optimized to obtain a heterogeneous feature vector. This may include the following steps:

[0076] Step S301: The chlorophyll concentration distribution matrix is ​​dynamically decomposed using spectral analysis to separate seasonal variation data reflecting long-term patterns and diurnal variation data reflecting short-term fluctuations.

[0077] Step S302: Based on preset hydrological characteristic indicators, extract corresponding subset data from seasonal variation data and diurnal variation data to obtain an initial feature vector.

[0078] Step S303: Construct a heterogeneous feature vector containing spatial heterogeneity dimension, temporal heterogeneity dimension and hydrological correlation dimension based on the initial feature vector, and obtain the distribution characteristic data of the heterogeneous feature vector through statistical analysis methods; the distribution characteristic data includes distribution mean, variance, extreme values ​​and distribution density.

[0079] Step S304: Using distribution characteristic data as input, and combining preset association rules for hydrological features and heterogeneity features, generate a dynamic adjustment model that is dynamically associated with hydrological features.

[0080] Step S305: Input the heterogeneous feature vector into the dynamic adjustment model for mapping processing to obtain the feature mapping result.

[0081] Step S306: Determine whether the feature mapping result is within the preset threshold range. If not, iteratively optimize the parameters of the dynamic adjustment model based on the deviation between the feature mapping result and the threshold range to determine the final heterogeneous feature vector.

[0082] Specifically, spectral analysis methods (such as Fourier spectral analysis and wavelet analysis) are used to dynamically decompose the previously obtained chlorophyll concentration distribution matrix. Based on the time scale characteristics corresponding to different frequency components (long-term frequencies correspond to seasonal scales, and short-term frequencies correspond to diurnal scales), seasonal variation data reflecting the long-term variation pattern of chlorophyll concentration and diurnal variation data reflecting short-term fluctuation characteristics are accurately separated. Based on preset hydrological characteristic indicators (including but not limited to key hydrological parameters such as water level, flow velocity, water depth, and turbidity), correlation coefficients between each hydrological indicator and seasonal and diurnal variation data are calculated using correlation analysis methods. Data sets with absolute correlation coefficient values ​​higher than preset thresholds are selected as corresponding subset data. Based on this subset data, an initial feature vector that can preliminarily characterize the relationship between chlorophyll concentration and hydrological characteristics is constructed. Based on the initial feature vector, a three-dimensional heterogeneous feature vector is further constructed, which includes spatial heterogeneity (reflecting the spatial differences in chlorophyll concentration in different areas of the water body), temporal heterogeneity (reflecting the temporal fluctuations in chlorophyll concentration at different times), and hydrological correlation (reflecting the correlation strength between chlorophyll concentration and various hydrological indicators). The heterogeneous feature vector is then quantified using descriptive statistical analysis methods to extract distribution characteristic data such as distribution mean, variance, extreme values, and distribution density, which can comprehensively reflect the distribution law of the feature vector.

[0083] Using the distribution characteristic data as input parameters, and combining the association rules of hydrological features and heterogeneous features established based on historical hydrological data and chlorophyll concentration monitoring data, a dynamic adjustment model dynamically associated with hydrological features is generated through algorithmic modeling. The constructed heterogeneous feature vector is input into the dynamic adjustment model for mapping transformation processing to obtain quantified feature mapping results. It is then determined whether the feature mapping results are within a preset threshold range (this threshold range is calibrated based on statistical analysis of historical hydrological-chlorophyll concentration association data). If the feature mapping results exceed the threshold range, the deviation value between the feature mapping results and the threshold range is calculated. Based on the magnitude and direction of the deviation value, the adjustment amplitude and direction of the dynamic adjustment model parameters are determined, and the model parameters are iteratively optimized. The heterogeneous feature vector is input into the optimized model again for mapping processing. The above judgment and optimization process is repeated until the feature mapping results meet the threshold requirements, and finally, the heterogeneous feature vector that accurately represents the core influencing factors of chlorophyll concentration is determined.

[0084] This embodiment achieves precise decomposition of the dynamic components of chlorophyll concentration at different time scales through spectral analysis, clarifying the independent influence of seasonal and diurnal variation patterns on chlorophyll concentration, laying a foundation for subsequent targeted analysis. The construction of three-dimensional heterogeneity feature vectors comprehensively covers the three core influence dimensions of space, time, and hydrology, avoiding the limitations of single-dimensional feature representation and ensuring the integrity of feature information. The extraction of distribution characteristic data provides quantitative support for the dynamic adjustment model, enabling the model to accurately optimize feature vectors based on objective data. The dynamic adjustment model, combined with association rules and iterative optimization mechanisms, effectively corrects the adaptation deviation between feature vectors and hydrological features, significantly improving the accuracy and relevance of heterogeneous feature vectors. The resulting heterogeneous feature vectors can deeply explore the intrinsic correlation between chlorophyll concentration and spatiotemporal heterogeneity and hydrological features.

[0085] In one embodiment, such as Figure 2 As shown, the influence weights of environment and climate are calculated based on a pure chlorophyll concentration signal sequence, and the chlorophyll concentration sequence is corrected. An effectiveness score is calculated based on the corrected chlorophyll concentration sequence. If the effectiveness score reaches a preset threshold, an optimized spatiotemporal distribution representation of chlorophyll concentration is generated. This process may include the following steps:

[0086] Step S401: Based on the pure chlorophyll concentration signal sequence combined with the carried geographic coordinates and timestamp information, extract the corresponding geographic environment difference factors and climate condition change factors.

[0087] Step S402: The correlation degree between geographical environment difference factors, climate condition change factors and pure chlorophyll concentration signal is determined by correlation analysis. Based on the correlation degree, a weighted fusion algorithm is used to calculate the influence weight of geographical environment difference and the influence weight of climate condition change, and the weight set is obtained by integration.

[0088] Step S403: Based on the distribution characteristics of each weight in the weight set and combined with the spatial fluctuation pattern of the pure chlorophyll concentration signal sequence, a spatial clustering algorithm is used to extract the local interference pattern of the high interference area. The geographical coordinate range of the high interference area is determined by pattern matching to obtain the high interference area identifier.

[0089] Step S404: For high-interference areas, identify the corresponding regions, combine the influence coefficients of each weight in the weight set on chlorophyll concentration, and use the weighted correction formula to calculate the concentration correction value of the corresponding region; for non-high-interference areas, directly use the pure chlorophyll concentration signal to integrate and obtain the corrected chlorophyll concentration sequence.

[0090] Step S405: The values ​​of each spatiotemporal node in the corrected chlorophyll concentration sequence are compared with the real-time water sample data with the corresponding geographic coordinates and timestamps. The comparison difference values ​​of each spatiotemporal node are obtained. Based on the comparison difference values, a comparison difference distribution covering the entire water area is constructed. The comparison difference distribution includes a spatial distribution map and a time variation curve.

[0091] Step S406: Calculate the mean, standard deviation, maximum deviation, and percentage of qualified nodes based on the comparison difference distribution, and use a comprehensive scoring model to obtain the effectiveness score of the interference separation strategy; qualified nodes are nodes whose difference values ​​are within the preset allowable error range.

[0092] Step S407: If the effectiveness score is lower than the preset effectiveness score threshold, then based on the weight contribution of the high deviation area in the comparison difference distribution, the weights of each influence in the weight set are iteratively adjusted until the effectiveness score reaches the effectiveness score threshold, and the optimized spatiotemporal distribution representation of chlorophyll concentration is obtained; the effectiveness score threshold is calibrated based on historical verification data and water quality monitoring accuracy requirements.

[0093] Specifically, based on the geographic coordinates and timestamp information carried by the pure chlorophyll concentration signal sequence, corresponding geographic environmental difference factors (including topographic slope, shoreline type, surrounding land use type, etc.) and climate condition change factors (including temperature, precipitation, sunshine duration, wind speed, etc.) are extracted. Pearson correlation analysis is used to calculate the correlation between each factor and the pure chlorophyll concentration signal. Based on this correlation, a "correlation degree-information entropy" weighted fusion algorithm is used to calculate the weights of geographic environmental difference influence and climate condition change influence, combining the comprehensive information entropy of each factor set to form a weight set. Based on the distribution characteristics of each weight in the weight set (such as distribution mean and variance), combined with the spatial fluctuation pattern of the pure chlorophyll concentration signal sequence (quantified by spatial coefficient of variation), spatial clustering algorithms (such as DBSCAN, K-means) are used to extract local interference patterns in high-interference areas. Pattern matching is then used to associate these patterns with geographic coordinates to define the geographic coordinate range of high-interference areas, generating high-interference area identifiers. For regions identified as high-interference areas, a weighted correction formula is used to calculate the concentration correction value for that region, combining the influence weights in the weight set and the preset effect coefficients. For non-high-interference areas, the pure chlorophyll concentration signal is directly used, and the data from both types of regions are integrated to obtain the corrected chlorophyll concentration sequence. The values ​​of each spatiotemporal node in the corrected chlorophyll concentration sequence are compared with the real-time water sample data of the corresponding geographic coordinates and timestamps point by point to obtain the comparison difference value of each spatiotemporal node. Based on this difference value, a comparison difference distribution covering the entire water area is constructed. This distribution includes a spatial distribution map reflecting spatial differences and a temporal change curve reflecting temporal changes. Based on the comparison difference distribution, the mean, standard deviation, maximum deviation, and percentage of qualified nodes (qualified nodes are nodes whose difference values ​​are within the preset allowable error range) are calculated. These indicators are input into a comprehensive scoring model (weighted summation of each indicator's weight) to obtain an effectiveness score for the interference separation strategy. A preset validity score threshold (calibrated based on historical validation data and water quality monitoring accuracy requirements) is set. If the validity score is lower than the threshold, the weight contribution of high deviation areas in the difference distribution is analyzed and compared. The values ​​of each influencing weight in the weight set are adjusted according to the contribution size. The concentration correction, difference comparison and scoring steps are re-executed. The iteration optimization is carried out until the validity score reaches the threshold, and finally the optimized spatiotemporal distribution representation of chlorophyll concentration is obtained.

[0094] This embodiment uses the pure chlorophyll concentration signal sequence as the core data source. Through multi-dimensional influencing factor extraction, it achieves comprehensive coverage of the driving factors of chlorophyll concentration changes. Combined with a weighted fusion algorithm of correlation and information entropy, it accurately quantifies the influence weights of geographical environment and climate conditions, providing a quantitative basis for concentration correction. Spatial clustering algorithms and pattern matching are used to accurately locate high-interference areas, enabling targeted concentration correction and avoiding over-adjustment of non-interference areas, thus improving the relevance and accuracy of the concentration sequence. A difference distribution is constructed based on point-by-point comparison of real-time water sample data, comprehensively characterizing the deviation features between the correction results and the measured data. Combined with a comprehensive score of multi-dimensional difference indicators, it achieves an objective evaluation of the effectiveness of the interference separation strategy. An iterative optimization mechanism dynamically adjusts the influence weights through weight contribution analysis of high-deviation areas, ensuring that the final output spatiotemporal distribution of chlorophyll concentration meets the preset accuracy requirements.

[0095] In one embodiment, the weights of the impact of geographical environmental differences and the impact of climate condition changes can be calculated using the following formula:

[0096]

[0097]

[0098] in, This indicates the weighting of the impact of geographical environmental differences. This indicates the weight of the impact of climate change. This represents the weighting balance coefficient. Pre-marked according to the type of lake or reservoir. The average Pearson correlation coefficient represents the set of geographical environmental difference factors and the pure chlorophyll concentration signal. , This indicates the number of factors contributing to geographical environmental differences. Indicates the first The correlation coefficients between geographical environmental factors and pure chlorophyll concentration signals The average Pearson correlation coefficient between the set of climate condition change factors and the pure chlorophyll concentration signal. , This indicates the number of factors affecting climate change. Indicates the first The correlation coefficients between the climate factors and the pure chlorophyll concentration signal The comprehensive information entropy represents the set of factors contributing to geographical environmental differences. The comprehensive information entropy represents the set of factors affecting climate change.

[0099] This embodiment introduces a balance coefficient adapted to the type of lake and reservoir. It allows for preset values ​​based on the characteristics of different water bodies, improving the scenario adaptability of weight calculation; combined with the average Pearson correlation coefficient The correlation strength between geographical environment, climate conditions and pure chlorophyll concentration signals is quantified, while incorporating comprehensive information entropy. , This approach characterizes the information richness and uncertainty of two types of factor sets, achieving a synergistic consideration of correlation and information effectiveness, avoiding the one-sidedness of weight quantification by a single-dimensional indicator; and its fractional structure naturally satisfies... + The normalization requirement of 1 ensures the rationality and interpretability of the weight allocation. Overall, this formula can accurately quantify the influence weights of geographical environmental differences and climate condition changes on chlorophyll concentration, providing a scientific and reliable quantitative basis for subsequent concentration correction in high-interference areas and improving the accuracy of the entire analysis process, effectively ensuring the accuracy and credibility of the spatiotemporal variation analysis results of chlorophyll concentration.

[0100] In one embodiment, the concentration correction value can be calculated using the following formula:

[0101]

[0102] in, This indicates the concentration correction value. This represents the raw value of the pure chlorophyll concentration signal within the high-interference region. Indicates the geographical environment adaptability coefficient. , This represents the average slope of the target area. This represents the average slope of the entire body of water. , Indicates the climate adaptability coefficient. , The average temperature of the target area over the past 30 days indicates... This indicates the average temperature of the entire water body over the past 30 days. , Indicates the interference intensity coefficient in the high interference region. , The standard deviation of the local interference pattern in the target region is represented by the following: The standard deviation of the pure chlorophyll concentration signal across the entire water body is represented. , This indicates the weighting of the impact of geographical environmental differences. This indicates the weight of the impact of climate change.

[0103] The calculation formula in this embodiment uses the original value of the pure chlorophyll concentration signal in the high-interference area as a benchmark, incorporates the weights of the influence of precise quantified geographical environment differences and climate condition changes, and combines a geographical environment adaptation coefficient constructed based on the differences in topographic slope and average temperature between the target area and the entire water area. Climate adaptability coefficient and the interference intensity coefficient characterizing the regional interference intensity. This approach achieves precise alignment between the correction process and regional geographical and climatic characteristics, as well as the degree of interference. Through the synergistic effect of weights, adaptation coefficients, and interference intensity coefficients, it can specifically compensate for deviations in chlorophyll concentration signals in highly interfered areas, accurately restoring the true chlorophyll concentration levels of those areas. This provides a scientifically quantifiable implementation path for concentration correction in highly interfered areas, significantly improving the accuracy of the corrected chlorophyll concentration sequence.

[0104] In one embodiment, the optimized spatiotemporal distribution representation of chlorophyll concentration is divided into spatiotemporal distribution data using a differentiated monitoring grid and the location information of key monitoring points is extracted to obtain a set of chlorophyll concentration monitoring parameters. This may include the following steps:

[0105] Step S501: The optimized spatiotemporal distribution representation of chlorophyll concentration is represented by a differentiated monitoring grid division method to generate a differentiated monitoring grid covering the entire water area and the spatiotemporal distribution data of chlorophyll concentration for the corresponding grid units.

[0106] Step S502: Based on the differentiated monitoring grid and the spatiotemporal distribution data of chlorophyll concentration, combined with the coefficient of variation of chlorophyll concentration within the grid unit, spatial coverage requirements and monitoring cost constraints, extract the geographic coordinate information of key monitoring points to form a preliminary monitoring point dataset containing the location of the monitoring point, the grid unit to which it belongs, and the initial monitoring period.

[0107] Step S503: Integrate the preliminary monitoring point dataset with the mean chlorophyll concentration, concentration change rate, and concentration extreme value corresponding to each monitoring point to construct a dynamic monitoring parameter matrix.

[0108] Step S504: Principal component analysis is used to remove redundant parameters from the dynamic monitoring parameter matrix, retain parameters whose information contribution is higher than a preset threshold, and determine the core monitoring parameter set for chlorophyll concentration.

[0109] Step S505: Based on the concentration sensitivity of each monitoring point in the core monitoring parameter set, set grid densification rules for areas with high information gain of core monitoring parameters to generate extended monitoring grids.

[0110] Step S506: Interpolate the optimized spatiotemporal distribution representation of chlorophyll concentration using the extended monitoring grid to obtain refined concentration distribution feature data that includes concentration gradient, spatial clustering features, and time abrupt change nodes.

[0111] Step S507: Based on the refined concentration distribution characteristic data, the correlation coefficient method and spatiotemporal autocorrelation analysis are used to calculate the concentration correlation strength between key monitoring points at different time scales, and obtain the correlation strength matrix between monitoring points.

[0112] Step S508: If the correlation strength between monitoring points in the correlation strength matrix is ​​lower than the preset threshold, then the resolution, boundary division and densification rules of the differentiated monitoring grid are adjusted in combination with the high heterogeneity region distribution in the refined concentration distribution feature data to generate a chlorophyll concentration monitoring parameter set.

[0113] Specifically, a differentiated monitoring grid division method was adopted to represent the optimized spatiotemporal distribution of chlorophyll concentration. The grid resolution was set according to the spatial heterogeneity of chlorophyll concentration in each region (higher resolution for high-heterogeneity regions than for low-heterogeneity regions), generating a differentiated monitoring grid covering the entire water area and spatiotemporal distribution data of chlorophyll concentration corresponding to each grid unit. Based on this differentiated monitoring grid and the spatiotemporal distribution data of chlorophyll concentration in each grid unit, combined with the coefficient of variation of chlorophyll concentration within the grid unit (quantifying the degree of spatial fluctuation), preset spatial coverage requirements, and actual monitoring cost constraints, a multi-objective screening model was used to extract the geographic coordinate information of key monitoring points, forming a preliminary monitoring point dataset containing the location of the monitoring point, its grid unit, and the initial monitoring period. The preliminary monitoring point dataset was integrated with indicators such as the mean chlorophyll concentration, concentration change rate, and concentration extreme values ​​corresponding to each monitoring point to construct a dynamic monitoring parameter matrix with the dimension of "monitoring point - monitoring indicator - time series". Principal component analysis (PCA) was used to reduce the dimensionality of the dynamic monitoring parameter matrix, calculate the information contribution of each parameter, and remove redundant parameters with information contributions below a preset threshold, retaining the core effective parameters to determine the core monitoring parameter set for chlorophyll concentration. Based on the concentration sensitivity of each monitoring point in the core monitoring parameter set, grid densification rules (such as increasing grid resolution and monitoring point density) were set in areas with high information gain of the core monitoring parameters to generate an extended monitoring grid. The extended monitoring grid was used to interpolate the optimized spatiotemporal distribution representation of chlorophyll concentration, filling the gaps between grid cells to obtain refined concentration distribution feature data including concentration gradients, spatial clustering features, and temporal abrupt change nodes. Based on the refined concentration distribution feature data, Pearson correlation coefficient method and spatiotemporal autocorrelation analysis (Moran's I index) were used to calculate the concentration correlation strength between key monitoring points at different time scales such as daily, weekly, and monthly, constructing a correlation strength matrix between monitoring points. A threshold for the correlation strength between monitoring points is preset (calibrated based on the requirements of water area hydrological connectivity and monitoring data validity). If the correlation strength between monitoring points in the correlation strength matrix is ​​lower than the threshold, the resolution, boundary division and densification rules of the differentiated monitoring grid are adjusted in combination with the distribution location of highly heterogeneous areas in the refined concentration distribution characteristic data. The monitoring points are then re-extracted and the core monitoring parameters are integrated to finally generate a chlorophyll concentration monitoring parameter set.

[0114] This embodiment achieves precise matching between monitoring resources and regional concentration heterogeneity through differentiated monitoring grid division, avoiding the problems of insufficient monitoring in highly heterogeneous areas or redundant resources in low-heterogeneity areas caused by uniform grid division. Preliminary monitoring points are extracted by combining the coefficient of variation, spatial coverage, and cost constraints, balancing monitoring accuracy and practical feasibility. The application of principal component analysis effectively eliminates redundant parameters, improving the relevance and effectiveness of the core monitoring parameter set. The construction and interpolation processing of the extended monitoring grid further refine the concentration distribution characteristics, providing high-precision data support for monitoring point correlation strength analysis. Through correlation strength calculation at different time scales, weak links in the monitoring point layout can be accurately identified, and grid parameters can be adjusted based on the distribution of highly heterogeneous areas to achieve dynamic optimization of the monitoring point layout. The generated chlorophyll concentration monitoring parameter set ensures both comprehensive monitoring coverage and data validity while also considering reasonable monitoring costs, providing a scientific and feasible implementation basis for long-term dynamic monitoring of chlorophyll concentration in lakes and reservoirs, effectively improving the accuracy and efficiency of water quality monitoring.

[0115] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0116] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method for analyzing the spatiotemporal variation of chlorophyll concentration in lakes and reservoirs based on multi-source remote sensing as described above.

[0117] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0118] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0119] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for analyzing the spatiotemporal variation of chlorophyll concentration in lakes and reservoirs based on multi-source remote sensing, characterized in that, The method includes: Multispectral images and real-time water sample data of lakes and reservoirs are collected. Chlorophyll concentration of the real-time water sample data is detected to obtain the measured chlorophyll concentration of the water sample carrying geographic coordinates and timestamps. A convolutional neural network inversion model is trained based on the multi-dimensional spectral feature vectors corresponding to geographic coordinates and the measured values ​​of chlorophyll concentration in the water sample. The multi-spectral images are then input into the trained inversion model to obtain the chlorophyll concentration distribution matrix. The chlorophyll concentration distribution matrix is ​​decomposed into dynamic components corresponding to seasonal and diurnal variations using spectral analysis. Subset data related to hydrological characteristics are extracted from the decomposed dynamic components, and the subset data is adjusted and optimized to obtain a heterogeneous feature vector. Determine whether the sediment suspension disturbance component in the heterogeneity feature vector exceeds a preset threshold. If so, filter the interference signal of the subset data using a machine learning classification method to obtain a pure chlorophyll concentration signal sequence. The influence weights of environment and climate are calculated based on the pure chlorophyll concentration signal sequence, and the chlorophyll concentration sequence is corrected. An effectiveness score is calculated based on the corrected chlorophyll concentration sequence. If the effectiveness score reaches a preset threshold, an optimized spatiotemporal distribution representation of chlorophyll concentration is generated. The optimized spatiotemporal distribution of chlorophyll concentration is represented by dividing the spatiotemporal distribution data using a differentiated monitoring grid and extracting the location information of key monitoring points to obtain a set of chlorophyll concentration monitoring parameters.

2. The method according to claim 1, characterized in that, The process involves training a convolutional neural network inversion model based on the multi-dimensional spectral feature vectors corresponding to geographic coordinates and the measured chlorophyll concentration of the water sample. The multi-spectral image is then input into the trained inversion model to obtain a chlorophyll concentration distribution matrix, including: Based on the spatiotemporal heterogeneity of the lake and reservoir water areas, pixel-level spectral information is extracted from the multi-spectral images to construct a multi-dimensional spectral feature vector that corresponds one-to-one with the geographic coordinates. Using the multi-dimensional spectral feature vector as input features and the measured chlorophyll concentration of the water sample as labels, a preset convolutional neural network model is trained to obtain a converged chlorophyll concentration inversion model. The full-pixel spectral information of the multi-band image is input into the chlorophyll concentration inversion model, and the model inference outputs a preliminary estimate of the chlorophyll concentration corresponding to the geographic coordinates of each pixel and the image timestamp. A spatiotemporal data fusion algorithm was used to interpolate and complete the preliminary estimates of chlorophyll concentration corresponding to adjacent timestamps, resulting in a spatiotemporally continuous chlorophyll concentration distribution matrix covering the entire water area.

3. The method according to claim 1, characterized in that, The chlorophyll concentration distribution matrix is ​​decomposed using spectral analysis to identify dynamic components corresponding to seasonal and diurnal variations. From these decomposed dynamic components, subset data related to hydrological characteristics are extracted and optimized to obtain a heterogeneous feature vector, including: The chlorophyll concentration distribution matrix was dynamically decomposed using spectral analysis to separate seasonal variation data reflecting long-term patterns and diurnal variation data reflecting short-term fluctuations. Based on preset hydrological characteristic indicators, corresponding subset data are extracted from the seasonal variation data and diurnal variation data to obtain an initial feature vector; Based on the initial feature vector, a heterogeneous feature vector containing spatial heterogeneity, temporal heterogeneity, and hydrological correlation dimensions is constructed. The distribution characteristic data of the heterogeneous feature vector is obtained through statistical analysis methods. The distribution characteristic data includes the distribution mean, variance, extreme values, and distribution density. Using the aforementioned distribution characteristic data as input, and combining it with preset association rules for hydrological features and heterogeneity features, a dynamic adjustment model that is dynamically associated with hydrological features is generated. The heterogeneous feature vectors are input into the dynamic adjustment model for mapping processing to obtain the feature mapping results; Determine whether the feature mapping result is within a preset threshold range. If not, iterate and optimize the parameters of the dynamic adjustment model based on the deviation between the feature mapping result and the threshold range to determine the final heterogeneous feature vector.

4. The method according to claim 1, characterized in that, The process involves calculating the influence weights of the environment and climate based on the purified chlorophyll concentration signal sequence and correcting the chlorophyll concentration sequence. An effectiveness score is then calculated based on the corrected chlorophyll concentration sequence. If the effectiveness score reaches a preset threshold, an optimized spatiotemporal distribution representation of chlorophyll concentration is generated, including: Based on the pure chlorophyll concentration signal sequence combined with the carried geographic coordinates and timestamp information, the corresponding geographic environment difference factors and climate condition change factors are extracted. The correlation between the geographical environment difference factor, the climate condition change factor and the pure chlorophyll concentration signal is determined by correlation analysis. Based on the correlation, a weighted fusion algorithm is used to calculate the influence weight of geographical environment difference and the influence weight of climate condition change, and the weight set is obtained by integration. Based on the distribution characteristics of each weight in the weight set, combined with the spatial fluctuation pattern of the pure chlorophyll concentration signal sequence, a spatial clustering algorithm is used to extract the local interference pattern of the high interference region. The geographical coordinate range of the high interference region is determined by pattern matching to obtain the high interference region identifier. For the regions corresponding to the high-interference regions, the concentration correction value of the corresponding regions is calculated by combining the influence coefficients of each weight in the weight set on the chlorophyll concentration using a weighted correction formula; for the non-high-interference regions, the pure chlorophyll concentration signal is directly used to integrate and obtain the corrected chlorophyll concentration sequence. The values ​​of each spatiotemporal node in the corrected chlorophyll concentration sequence are compared with the real-time water sample data with the corresponding geographic coordinates and timestamps. The comparison difference values ​​of each spatiotemporal node are obtained. Based on the comparison difference values, a comparison difference distribution covering the entire water area is constructed. The comparison difference distribution includes a spatial distribution map and a temporal variation curve. Based on the comparison difference distribution, the mean difference, standard deviation, maximum deviation, and percentage of qualified nodes are calculated. A comprehensive scoring model is used to obtain the effectiveness score of the interference separation strategy. The qualified nodes are those whose difference values ​​are within the preset allowable error range. If the effectiveness score is lower than the preset effectiveness score threshold, then based on the weight contribution of the high deviation region in the comparison difference distribution, the weights of each influence in the weight set are iteratively adjusted until the effectiveness score reaches the effectiveness score threshold, and the optimized spatiotemporal distribution representation of chlorophyll concentration is obtained; the effectiveness score threshold is calibrated based on historical verification data and water quality monitoring accuracy requirements.

5. The method according to claim 4, characterized in that, The weights of the influence of geographical environment differences and the influence of climate condition changes are calculated using the following formulas: in, This indicates the weighting of the impact of geographical environmental differences. This indicates the weight of the impact of climate change. This represents the weighting balance coefficient. Pre-marked according to the type of lake or reservoir. The average Pearson correlation coefficient represents the set of geographical environmental difference factors and the pure chlorophyll concentration signal. , This indicates the number of factors contributing to geographical environmental differences. Indicates the first The correlation coefficients between geographical environmental factors and pure chlorophyll concentration signals The average Pearson correlation coefficient between the set of climate condition change factors and the pure chlorophyll concentration signal. , This indicates the number of factors affecting climate change. Indicates the first The correlation coefficients between the climate factors and the pure chlorophyll concentration signal The comprehensive information entropy represents the set of factors contributing to geographical environmental differences. The comprehensive information entropy represents the set of factors affecting climate change.

6. The method according to claim 4, characterized in that, The concentration correction value is calculated using the following formula: in, This indicates the concentration correction value. This represents the raw value of the pure chlorophyll concentration signal within the high-interference region. Indicates the geographical environment adaptability coefficient. , This represents the average slope of the target area. This represents the average slope of the entire body of water. , Indicates the climate adaptability coefficient. , The average temperature of the target area over the past 30 days indicates... This indicates the average temperature of the entire water body over the past 30 days. , Indicates the interference intensity coefficient in the high interference region. , The standard deviation of the local interference pattern in the target region is represented by the following: The standard deviation of the pure chlorophyll concentration signal across the entire water body is represented. , This indicates the weighting of the impact of geographical environmental differences. This indicates the weight of the impact of climate change.

7. The method according to claim 1, characterized in that, The optimized spatiotemporal distribution representation of chlorophyll concentration utilizes a differentiated monitoring grid to divide the spatiotemporal distribution data and extracts the location information of key monitoring points to obtain a set of chlorophyll concentration monitoring parameters, including: The optimized spatiotemporal distribution of chlorophyll concentration is represented by a differentiated monitoring grid division method to generate a differentiated monitoring grid covering the entire water area and the spatiotemporal distribution data of chlorophyll concentration for the corresponding grid units; Based on the differentiated monitoring grid and the spatiotemporal distribution data of chlorophyll concentration, combined with the coefficient of variation of chlorophyll concentration within the grid unit, spatial coverage requirements and monitoring cost constraints, the geographic coordinate information of key monitoring points is extracted to form a preliminary monitoring point dataset containing the location of the monitoring point, the grid unit to which it belongs, and the initial monitoring period. The preliminary monitoring point dataset is integrated with the mean chlorophyll concentration, concentration change rate, and concentration extreme value corresponding to each monitoring point to construct a dynamic monitoring parameter matrix. Principal component analysis is used to remove redundant parameters from the dynamic monitoring parameter matrix, retaining parameters whose information contribution is higher than a preset threshold, to determine the core monitoring parameter set for chlorophyll concentration; Based on the concentration sensitivity of each monitoring point in the core monitoring parameter set, a grid densification rule is set for areas with high information gain of the core monitoring parameters to generate an extended monitoring grid. The optimized spatiotemporal distribution representation of chlorophyll concentration is interpolated using the extended monitoring grid to obtain refined concentration distribution feature data that includes concentration gradient, spatial clustering features, and temporal abrupt change nodes. Based on the refined concentration distribution characteristic data, the correlation coefficient method and spatiotemporal autocorrelation analysis are used to calculate the concentration correlation strength between key monitoring points at different time scales, and the correlation strength matrix between monitoring points is obtained. If the correlation strength between monitoring points in the correlation strength matrix is ​​lower than a preset threshold, then the resolution, boundary division, and encryption rules of the differentiated monitoring grid are adjusted based on the high heterogeneity region distribution in the refined concentration distribution feature data to generate a chlorophyll concentration monitoring parameter set.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images