A method for intraday dynamic prediction of cyanobacterial blooms driven by meteorological and water quality synergy.
By combining high temporal resolution data from geostationary meteorological satellites with water quality parameters and employing a multi-iteration BP neural network optimization method, cyanobacterial bloom regions were identified and a prediction model was established. This solved the problem of strong dependence on meteorological factors in existing models, enabling accurate prediction and efficient management of intraday dynamics of cyanobacterial blooms.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- NAT SATELLITE METEOROLOGICAL CENT
- Filing Date
- 2025-08-27
- Publication Date
- 2026-05-26
AI Technical Summary
Existing cyanobacterial bloom prediction models rely too heavily on meteorological factors and neglect the background accumulation effect of water quality parameters and the synergistic regulation mechanism between meteorological factors and water quality parameters, resulting in limited prediction accuracy and decreased reliability, especially when water quality monitoring is lacking, the prediction results are inaccurate.
We employed a multi-iteration iterative method to find the optimal initialization of weights and thresholds for a BP neural network. Combining high temporal resolution data from geostationary meteorological satellites and water quality parameters, we identified cyanobacterial bloom regions using a multi-index consensus method. We then constructed a BP neural network model and optimized the number of neurons in the hidden layer to improve prediction accuracy.
It achieves accurate intraday prediction of cyanobacterial bloom dynamics, reduces mean absolute error, improves the model's generalization ability, supports rolling predictions for the next 24 hours, and provides a decision-making window for emergency prevention and control of cyanobacterial blooms.
Smart Images
Figure CN121189539B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of water environment monitoring, specifically to a method for intraday dynamic prediction of cyanobacterial blooms based on meteorological-water quality synergistic driving. Background Technology
[0002] Cyanobacterial blooms, a serious problem in global aquatic ecosystems, not only have a profound impact on the natural environment but also pose an increasingly serious threat to human health and socio-economic development. Their highly toxic metabolites and disruption of aquatic ecological balance have made them a hot topic in contemporary environmental science research. The formation of cyanobacterial blooms involves multiple complex factors, among which meteorological and water quality conditions play a crucial role in their occurrence and development.
[0003] Studies have shown that water quality parameters, including nutrients and chlorophyll a concentration, are the material basis for cyanobacterial blooms, directly determining the potential for algal biomass accumulation in water bodies. Meteorological factors such as air temperature, light intensity, and wind speed, on the other hand, regulate algal physiological activities and apparent aggregation states, triggering the spatiotemporal dynamics of blooms. For eutrophic lakes, long-term high nutrient levels have already laid the biomass foundation for blooms, making them highly susceptible to being triggered by short-term suitable meteorological conditions. Ignoring the background role of water quality parameters and relying solely on meteorological factors to build predictive models will result in limited accuracy. Therefore, integrating the synergistic driving mechanisms of water quality parameters and meteorological factors to construct cyanobacterial bloom prediction models is crucial for achieving accurate early warning and tiered management.
[0004] Current research on predicting cyanobacterial blooms in lakes mainly includes physical models based on algal growth and inductive models based on historical data. Physical models based on algal growth rely on the close response relationship between blooms and environmental factors. However, bloom processes are complex, exhibiting significant randomness and uncertainty. Analyzing the impact of environmental factors on blooms from an ecological perspective makes model construction complex and accurate prediction difficult. Inductive models based on historical data provide a powerful tool for time-series prediction of target variables. Therefore, domestic and international experts and scholars often utilize regression analysis or correlation analysis to establish meteorological prediction models for cyanobacterial blooms. Representative models include simple regression models, decision tree models, genetic algorithm models, and artificial neural network models. Among these, artificial neural network models, due to their powerful nonlinear modeling capabilities, can better handle nonlinear relationships, approximate arbitrary continuous functions, and simulate the interactions between multiple factors and their impact on the target factor, thus improving prediction accuracy and gradually becoming a research hotspot. Applying artificial neural network models to learn patterns and regularities in historical data can more accurately predict the occurrence of cyanobacterial blooms.
[0005] However, the influence of meteorological factors and water quality parameters on cyanobacterial blooms typically exhibits a non-linear variation pattern, limiting the application of traditional statistical methods for quantitative analysis of the causes of cyanobacterial blooms at different stages. Influenced by environmental changes and the vertical migration of cyanobacteria, the spatiotemporal dynamics of cyanobacterial blooms change dramatically within a single day. Existing cyanobacterial bloom prediction models, based on Landsat and MODIS data to identify cyanobacteria and establish prediction models, have lower temporal resolution compared to geostationary meteorological satellites, failing to capture the drastic changes in cyanobacterial blooms within a single day. Most existing models over-rely on meteorological factors, neglecting the decisive role of water quality parameters in the long-term accumulation of cyanobacterial biomass, and their synergistic regulatory mechanisms with meteorological factors. This not only limits the accuracy of model predictions to the one-sidedness of meteorological factors, but also significantly reduces the reliability of prediction results due to the lack of key variables when water quality monitoring is unavailable. Summary of the Invention
[0006] In view of this, the purpose of this invention is to propose a method for intraday dynamic prediction of cyanobacterial blooms based on meteorological-water quality synergy. This method utilizes multiple iterations to find the optimal initialization method for the weights and thresholds of a BP neural network, training and verifying its accuracy by gradually increasing the number of neurons in the hidden layer. It fully leverages the high temporal resolution of geostationary meteorological satellites to identify cyanobacterial blooms and establish a prediction model, thereby achieving intraday prediction of the cyanobacterial bloom area.
[0007] To achieve the above objectives, the present invention provides the following technical solution:
[0008] To achieve the above objectives, in a first aspect, the present invention provides a method for intraday dynamic prediction of cyanobacterial blooms based on meteorological-water quality synergistic driving, comprising the following steps:
[0009] Based on satellite data, the hourly cyanobacterial bloom area results images were identified using the multi-index consensus method, and the actual coverage area of the cyanobacterial bloom was calculated.
[0010] A BP neural network model was constructed and optimized using combined meteorological factors and water quality parameters as input and the actual coverage area of cyanobacterial blooms as output.
[0011] The optimized model is used to predict the estimated coverage area of cyanobacterial blooms hourly in the future, and the efficiency coefficient between the estimated coverage area and the actual coverage area of cyanobacterial blooms is calculated.
[0012] As a further aspect of the present invention, the satellite data is geostationary satellite data with a time resolution of ≤1 hour obtained based on FY4A / AGRI; and cyanobacterial blooms are identified hourly on a daily basis based on the satellite data.
[0013] As a further aspect of the present invention, the method of identifying hourly cyanobacterial bloom area results images using a multi-index consensus method and calculating the actual coverage area of the cyanobacterial bloom includes the following steps:
[0014] Radiometric calibration, geometric correction, and atmospheric correction were performed on FY4A / AGRI satellite data, and the data was clipped using lake boundary vectors to obtain preprocessed satellite data.
[0015] The HOT index is calculated based on preprocessed satellite data, wherein the HOT index is derived from the blue and red light bands, i.e.:
[0016] ;
[0017] In the formula, Indicates the reflectivity of the blue light band. The red light band reflectivity is represented; a threshold 'a' is selected to perform threshold segmentation on the HOT index, and visual interpretation is combined to improve cloud recognition accuracy and obtain image data after eliminating cloud effects.
[0018] Based on satellite imagery data processed by cloud recognition, cyanobacterial blooms were extracted using Normalized Difference Vegetation Index (NDVI), Ratio Vegetation Index (RVI), and Difference Vegetation Index (DVI), respectively. The calculation formulas are as follows:
[0019] ;
[0020] ;
[0021] ;
[0022] In the formula, Indicates near-infrared reflectivity. Representing the reflectivity in the red light band, thresholds b, c, and d were selected to perform threshold segmentation on the three indices to extract cyanobacterial blooms, and visual interpretation was combined to improve the extraction accuracy.
[0023] The actual coverage area of cyanobacterial blooms was calculated hourly based on the images obtained from the cyanobacterial bloom identification results.
[0024] As a further aspect of the present invention, the preliminary results of cyanobacterial blooms are extracted using Normalized Difference Vegetation Index (NDVI), Ratio Vegetation Index (RVI), and Difference Vegetation Index (DVI), respectively. The final hourly cyanobacterial bloom identification results are then determined using a multi-index consensus method, wherein the multi-index consensus method formula is:
[0025] ;
[0026] ;
[0027] ;
[0028] ;
[0029] Where i represents the pixel number in the image. , and This represents the binarization result of pixel i under NDVI, RVI and DVI indices, where 1 represents cyanobacterial bloom and 0 represents non-cyanobacterial bloom; This represents the final result identified by the multi-index consensus method for pixel i, where 1 represents cyanobacterial bloom and 0 represents non-cyanobacterial bloom.
[0030] As a further aspect of the present invention, the formula for calculating the actual coverage area of the cyanobacterial bloom is as follows:
[0031] ;
[0032] ;
[0033] Where n is the total number of pixels affected by cyanobacterial blooms, and i is the pixel number affected by cyanobacterial blooms. Let i be the area of the i-th cyanobacterial bloom pixel. This represents the actual area covered by cyanobacterial blooms. Let i be the cyanobacterial bloom coverage index for the i-th pixel. This represents the normalized vegetation index of the i-th pixel. and These are the NDVI values for clean water bodies and cyanobacterial blooms, typically taken as -0.2 and 0.81, respectively.
[0034] As a further aspect of the present invention, when constructing and optimizing the BP neural network model, the method further includes setting the number of hidden layer neurons to 5-30, generating 26,000 candidate models by randomly initializing parameters and thresholds, and selecting the optimal model with the goal of minimizing the mean absolute error (MAE). The formula for calculating the mean absolute error (MAE) is as follows:
[0035] ;
[0036] In the formula, The mean absolute error, Represents the actual observed value. This represents the predicted value.
[0037] As a further aspect of the present invention, the optimization of the BP neural network model includes the following steps:
[0038] a) Normalize the hourly data of actual coverage area of cyanobacterial blooms and corresponding meteorological factors and water quality parameters as sample data to obtain normalized sample data.
[0039] b) Input the normalized sample data into the BP neural network, randomly set the initial parameters and threshold, and set the number of hidden layer neurons to 5 to 30, and set the number of iterations to 1000 for each neuron's data volume;
[0040] c) Train the BP neural network, output the network's prediction results, and determine whether the maximum number of iterations has been reached. If not, return to step b). If so, increase the number of hidden layer neurons sequentially with a step size of 1, and determine whether the number of neurons exceeds the preset range. If so, complete the training; otherwise, return to step b). 26,000 network models with different numbers of hidden layer neurons and initial parameter settings are obtained through iterative training.
[0041] As a further aspect of the present invention, when calculating the efficiency coefficient between the estimated coverage area and the actual coverage area of cyanobacterial blooms, the method further includes:
[0042] By comparing the mean absolute error between the network model's prediction of the cyanobacterial bloom coverage area and the actual coverage area of the cyanobacterial bloom, the network model with the smallest mean absolute error is selected as the prediction model.
[0043] As a further aspect of the present invention, the efficiency coefficient for calculating the estimated coverage area of cyanobacterial blooms versus the actual coverage area of cyanobacterial blooms includes the following steps:
[0044] Based on different network models, the estimated coverage area of cyanobacterial blooms predicted by each model is output.
[0045] Calculate the average absolute error between the predicted cyanobacterial bloom coverage area and the actual cyanobacterial bloom coverage area of each network model.
[0046] The network model with the smallest mean absolute error was selected as the prediction model, and the efficiency coefficient between the predicted actual coverage area of cyanobacterial blooms and the actual coverage area of real cyanobacterial blooms was calculated.
[0047] As a further aspect of the present invention, the formula for calculating the efficiency coefficient is as follows:
[0048] ;
[0049] In the formula, E represents the efficiency coefficient. Represents the actual observed value. Indicates the predicted value. This represents the average of the actual observed values, where n is the total number of samples.
[0050] Compared with existing technologies, the intraday dynamic prediction method for cyanobacterial blooms based on meteorological-water quality synergy proposed in this invention has the following beneficial effects:
[0051] 1. This invention utilizes a multi-iteration method to find the optimal initialization method for the weights and thresholds of a BP neural network. It trains and verifies the accuracy by gradually increasing the number of neurons in the hidden layer. Simultaneously, it integrates the background accumulation effect of water quality parameters and the short-term triggering effect of meteorological factors, solving the problem of the one-sidedness of traditional models that over-rely on meteorological data. Through the meteorological-water quality dual-drive mechanism, it reduces the average absolute error of prediction, which is significantly better than a single meteorological-driven model. Furthermore, based on geostationary satellite data such as FY4A / AGRI, it overcomes the limitations of low-resolution data such as Landsat / MODIS. Combining the multi-index consensus method (NDVI+RVI+DVI) and the cyanobacterial coverage index (FCI), it achieves accurate tracking of drastic intraday changes in cyanobacterial blooms and reduces the error in calculating the actual biomass area.
[0052] 2. This invention fully utilizes the high temporal resolution of geostationary meteorological satellites to identify cyanobacterial blooms and establish a prediction model. It enables daily prediction of the actual coverage area of cyanobacterial blooms. By using a three-indice joint judgment (output only when at least two indices simultaneously identify cyanobacteria), it reduces the misjudgment rate caused by a single index or threshold, lowers the misjudgment rate due to cloud interference, and improves the accuracy of identifying mixed pixels at the bloom edge. Through iterative optimization of the number of neurons combined with 26,000 parameter initializations, it avoids local optima. The MAE minimization criterion ensures the model's generalization ability, supporting rolling predictions for the next 24 hours. This provides a decision-making window for emergency prevention and control of cyanobacterial blooms, better simulating the dynamic changes of cyanobacterial blooms, and providing a basis for the prevention and management of water environment governance and eutrophication control.
[0053] These or other aspects of this application will become more apparent from the following description of embodiments. It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the application. Attached Figure Description
[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the accompanying drawings used in the description of the exemplary embodiments or related technologies will be briefly introduced below. The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation thereof. In the drawings:
[0055] Figure 1 This is a flowchart of a method for intraday dynamic prediction of cyanobacterial blooms based on meteorological-water quality synergistic driving according to an embodiment of the present invention.
[0056] Figure 2This is a flowchart of neural network training in a method for predicting the intraday dynamics of cyanobacterial blooms based on meteorological-water quality synergy, according to an embodiment of the present invention. Detailed Implementation
[0057] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings, but it should be understood that the scope of protection of the present invention is not limited to the specific embodiments.
[0058] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.
[0059] This invention proposes a diurnal dynamic prediction method for cyanobacterial blooms based on meteorological and water quality synergy. It utilizes multiple iterations to find the optimal initialization method for the weights and thresholds of a BP neural network, training and verifying its accuracy by gradually increasing the number of neurons in the hidden layer. Taking full advantage of the high temporal resolution of geostationary meteorological satellites, it identifies cyanobacterial blooms and establishes a prediction model, achieving diurnal prediction of the cyanobacterial bloom area.
[0060] like Figure 1 and Figure 2 As shown, an embodiment of the present invention provides a method for intraday dynamic prediction of cyanobacterial blooms based on meteorological-water quality synergistic driving force. The method includes the following steps:
[0061] Step S10: Based on satellite data, identify hourly cyanobacterial bloom area results images using the multi-index consensus method, and calculate the actual coverage area of cyanobacterial blooms.
[0062] In this step, the satellite data is geostationary satellite data with a time resolution of ≤1 hour, acquired based on FY4A / AGRI; hourly cyanobacterial blooms are identified daily based on the satellite data. In some embodiments, the FY4A / AGRI satellite data can be replaced with other geostationary satellite data with red and near-infrared bands that can provide hourly imagery.
[0063] In this embodiment, the hourly cyanobacterial bloom area results image is identified using the multi-index consensus method, and the actual coverage area of the cyanobacterial bloom is calculated, including the following steps:
[0064] Step 1) Perform radiometric calibration, geometric correction, and atmospheric correction on the FY4A / AGRI satellite data, and clip it using lake boundary vectors to obtain preprocessed satellite data;
[0065] Step 2) Calculate the HOT index based on the preprocessed satellite data, wherein the HOT index is calculated using the blue and red light bands, i.e.:
[0066] ;
[0067] In the formula, Indicates the reflectivity of the blue light band. The red light band reflectivity is represented; a threshold 'a' is selected to perform threshold segmentation on the HOT index, and visual interpretation is combined to improve cloud recognition accuracy and obtain image data after eliminating cloud effects.
[0068] Step 3) Based on the satellite imagery data processed by cloud recognition, cyanobacterial blooms are extracted using Normalized Difference Vegetation Index (NDVI), Ratio Vegetation Index (RVI), and Difference Vegetation Index (DVI), respectively. The calculation formulas are as follows:
[0069] ;
[0070] ;
[0071] ;
[0072] In the formula, Indicates near-infrared reflectivity. Representing the reflectivity in the red light band, thresholds b, c, and d were selected to perform threshold segmentation on the three indices to extract cyanobacterial blooms, and visual interpretation was combined to improve the extraction accuracy.
[0073] Step 4) Calculate the actual coverage area of cyanobacterial blooms hourly based on the images of the cyanobacterial bloom identification results.
[0074] The formula for calculating the actual coverage area of the cyanobacterial bloom is as follows:
[0075] ;
[0076] ;
[0077] Where n is the total number of pixels affected by cyanobacterial blooms, and i is the pixel number affected by cyanobacterial blooms. Let i be the area of the i-th cyanobacterial bloom pixel. This represents the actual area covered by cyanobacterial blooms. Let i be the cyanobacterial bloom coverage index for the i-th pixel. This represents the normalized vegetation index of the i-th pixel. and These are the NDVI values for clean water bodies and cyanobacterial blooms, typically taken as -0.2 and 0.81, respectively.
[0078] In this embodiment, to overcome the uncertainty of a single index or threshold and improve the reliability of the identification results, a multi-index consensus method is used to determine the final daily hourly cyanobacterial bloom identification result image based on the preliminary extraction results of the three index methods. The preliminary results obtained from cyanobacterial blooms are extracted using the Normalized Difference Vegetation Index (NDVI), Ratio Vegetation Index (RVI), and Difference Vegetation Index (DVI), respectively. The multi-index consensus method is then used to determine the final daily hourly cyanobacterial bloom identification result image. The formula for the multi-index consensus method is:
[0079] ;
[0080] ;
[0081] ;
[0082] ;
[0083] Where i represents the pixel number in the image. , and This represents the binarization result of pixel i under NDVI, RVI and DVI indices, where 1 represents cyanobacterial bloom and 0 represents non-cyanobacterial bloom; This represents the final result identified by the multi-index consensus method for pixel i, where 1 represents cyanobacterial bloom and 0 represents non-cyanobacterial bloom.
[0084] Step S20: Using the combined data of meteorological factors and water quality parameters as input and the actual coverage area of cyanobacterial blooms as output, construct and optimize a BP neural network model.
[0085] In this step, see Figure 2 As shown, the construction and optimization of the BP neural network model also includes setting the number of hidden layer neurons to 5-30, generating 26,000 candidate models through random initialization of parameters and thresholds, and selecting the optimal model with the goal of minimizing the mean absolute error (MAE). The formula for calculating the mean absolute error (MAE) is as follows:
[0086] ;
[0087] In the formula, The mean absolute error, Represents the actual observed value. This represents the predicted value.
[0088] The optimization of the BP neural network model includes the following steps:
[0089] a) Normalize the hourly data of actual coverage area of cyanobacterial blooms and corresponding meteorological factors and water quality parameters as sample data to obtain normalized sample data.
[0090] b) Input the normalized sample data into the BP neural network, randomly set the initial parameters and threshold, and set the number of hidden layer neurons to 5 to 30, and set the number of iterations to 1000 for each neuron's data volume;
[0091] c) Train the BP neural network, output the network's prediction results, and determine whether the maximum number of iterations has been reached. If not, return to step b). If so, increase the number of hidden layer neurons sequentially with a step size of 1, and determine whether the number of neurons exceeds the preset range. If so, complete the training; otherwise, return to step b). 26,000 network models with different numbers of hidden layer neurons and initial parameter settings are obtained through iterative training.
[0092] In step S20, the hourly actual coverage area of cyanobacterial blooms and the corresponding combined data of meteorological factors and water quality parameters are used as sample data and normalized to obtain normalized sample data. The normalized sample data is then input into a backpropagation (BP) neural network, with initial parameters and thresholds randomly set. The number of hidden layer neurons is set to 5 to 30, and the number of iterations is set to 1000 for each neuron. The BP neural network is trained, and the network prediction results are output. It is then determined whether the maximum number of iterations has been reached. If not, the normalized sample data is input into the BP neural network. If so, the number of hidden layer neurons is increased sequentially with a step size of 1, and it is determined whether the number of neurons exceeds the preset range. If so, the training is completed; otherwise, the normalized sample data is input into the BP neural network. Through iterative training, 26,000 network models with different numbers of hidden layer neurons and initial parameter settings are obtained.
[0093] In some embodiments, the meteorological station data corresponding to the meteorological factors can be replaced with other published meteorological data products, and the meteorological data acquisition time is consistent with the satellite data acquisition time. This invention utilizes multiple iterations to find the optimal initialization method for the weights and thresholds of a BP neural network. By gradually increasing the number of neurons in the hidden layer, the invention trains and verifies its accuracy, providing a new approach for predicting cyanobacterial blooms based on BP neural networks and optimizing the performance of the BP neural network cyanobacterial prediction model. By fully utilizing the high temporal resolution of geostationary meteorological satellites, this method identifies cyanobacterial blooms and establishes a prediction model, enabling intraday prediction of the actual coverage area of cyanobacterial blooms. This method can better simulate the dynamic changes of cyanobacterial blooms, providing a basis for the prevention and management of water environment governance and eutrophication control.
[0094] Step S30: Use the optimized model to predict the estimated coverage area of cyanobacterial blooms hourly in the future, and calculate the efficiency coefficient between the estimated coverage area and the actual coverage area of cyanobacterial blooms.
[0095] This step, when calculating the efficiency coefficient between the estimated and actual coverage areas of cyanobacterial blooms, also includes:
[0096] By comparing the mean absolute error between the network model's prediction of the cyanobacterial bloom coverage area and the actual coverage area of the cyanobacterial bloom, the network model with the smallest mean absolute error is selected as the prediction model.
[0097] In this embodiment, the efficiency coefficient for calculating the estimated coverage area of cyanobacterial blooms versus the actual coverage area of cyanobacterial blooms includes the following steps:
[0098] Based on different network models, the estimated coverage area of cyanobacterial blooms predicted by each model is output.
[0099] Calculate the average absolute error between the predicted cyanobacterial bloom coverage area and the actual cyanobacterial bloom coverage area of each network model.
[0100] The network model with the smallest mean absolute error was selected as the prediction model, and the efficiency coefficient between the predicted actual coverage area of cyanobacterial blooms and the actual coverage area of real cyanobacterial blooms was calculated.
[0101] The formula for calculating the efficiency coefficient is as follows:
[0102] ;
[0103] In the formula, E represents the efficiency coefficient. Represents the actual observed value. Indicates the predicted value. This represents the average of the actual observed values, where n is the total number of samples.
[0104] This invention presents a diurnal dynamic prediction method for cyanobacterial blooms based on meteorological and water quality synergistic driving. It utilizes multiple iterations to find the optimal initialization method for the weights and thresholds of a BP neural network. The accuracy is tested by gradually increasing the number of neurons in the hidden layer. Simultaneously, it integrates the background accumulation effect of water quality parameters with the short-term triggering effect of meteorological factors, solving the problem of traditional models' over-reliance on meteorological data. The meteorological-water quality dual-driving mechanism reduces the mean absolute error of prediction, significantly outperforming single meteorological-driven models. Furthermore, based on geostationary satellite data such as FY4A / AGRI, it overcomes the limitations of low-resolution data such as Landsat / MODIS, combining a multi-index consensus method (NDVI+RVI+DVI) with the cyanobacterial coverage index (FCI) to accurately track drastic diurnal changes in cyanobacterial blooms, reducing errors in the calculation of actual biomass area.
[0105] This invention also fully utilizes the high temporal resolution of geostationary meteorological satellites to identify cyanobacterial blooms and establish a prediction model, enabling intraday prediction of the actual coverage area of cyanobacterial blooms. By using a three-index joint judgment (output is only given if at least two indices simultaneously identify cyanobacteria), the misjudgment rate caused by a single index or threshold is reduced, the misjudgment rate due to cloud interference is decreased, and the accuracy of identifying mixed pixels at the edge of blooms is improved. Through iterative optimization of the number of neurons combined with 26,000 parameter initializations, local optima are avoided, and the MAE minimization criterion ensures the model's generalization ability, supporting rolling predictions for the next 24 hours. This provides a decision-making window for emergency prevention and control of blooms, and can better simulate the dynamic changes of cyanobacterial blooms, providing a basis for the prevention and management of water environment governance and eutrophication control.
[0106] The above are exemplary embodiments disclosed in this invention. However, it should be noted that various changes and modifications can be made without departing from the scope of the embodiments of this invention as defined by the claims. The functions, steps, and / or actions of the methods according to the disclosed embodiments described herein do not need to be performed in any particular order. Furthermore, although the elements disclosed in the embodiments of this invention may be described or claimed individually, they may be understood as multiple unless explicitly limited to a singular number.
[0107] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of the invention (including the claims) is limited to these examples. Within the framework of the invention, technical features of the above embodiments or different embodiments can be combined, and many other variations of different aspects of the invention exist, which are not provided in the details for the sake of brevity. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the invention should be included within the protection scope of the invention.
[0108] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.
Claims
1. A method for intraday dynamic prediction of cyanobacterial blooms based on meteorological-water quality synergistic driving force, characterized in that, The method includes the following steps: Based on geostationary satellite data with a temporal resolution of ≤1 hour acquired by FY4A / AGRI, hourly cyanobacterial blooms were identified daily. Radiometric calibration, geometric correction, and atmospheric correction were performed on the FY4A / AGRI satellite data, and the data was cropped using lake boundary vectors to obtain preprocessed satellite data. The HOT index was calculated based on the preprocessed satellite data, wherein the HOT index was derived using blue and red light bands. Based on the cloud-recognition-processed satellite imagery data, cyanobacterial blooms were extracted using the Normalized Difference Vegetation Index (NDVI), Ratio Vegetation Index (RVI), and Difference Vegetation Index (DVI). The actual hourly coverage area of cyanobacterial blooms was calculated based on the identified images. Using combined meteorological factors and water quality parameters as input and the actual coverage area of cyanobacterial blooms as output, a BP neural network model is constructed and optimized. When constructing and optimizing the BP neural network model, the number of hidden layer neurons is set to 5-30, 26,000 candidate models are generated by randomly initializing parameters and thresholds, and the optimal model is selected with the goal of minimizing the mean absolute error. The optimized model is used to predict the estimated coverage area of cyanobacterial blooms hourly in the future. The efficiency coefficients of the estimated coverage area and the actual coverage area of cyanobacterial blooms are calculated. The mean absolute error between the network model's prediction of the estimated coverage area and the actual coverage area of cyanobacterial blooms is compared. The network model with the smallest mean absolute error is selected as the prediction model.
2. The intraday dynamic prediction method for cyanobacterial blooms based on meteorological-water quality synergistic driving as described in claim 1, characterized in that, The HOT index is calculated using the blue and red light bands, namely: ； In the formula, Indicates the reflectivity of the blue light band. Indicates the reflectivity in the red light band; Threshold a is selected to perform threshold segmentation on the HOT index, and visual interpretation is combined to improve cloud recognition accuracy and obtain image data after eliminating cloud effects. Cyanobacterial blooms were extracted using the Normalized Difference Vegetation Index (NDVI), Ratio Vegetation Index (RVI), and Difference Vegetation Index (DVI), respectively, and the calculation formulas are as follows: ； ； ； In the formula, Indicates near-infrared reflectivity. Representing the reflectivity in the red light band, thresholds b, c, and d were selected to perform threshold segmentation on the three indices to extract cyanobacterial blooms, and visual interpretation was combined to improve the extraction accuracy.
3. The intraday dynamic prediction method for cyanobacterial blooms based on meteorological-water quality synergistic driving as described in claim 2, characterized in that, Preliminary results of cyanobacterial bloom identification were extracted using Normalized Difference Vegetation Index (NDV), Ratio Vegetation Index (RDV), and Difference Vegetation Index (DVI). The final hourly cyanobacterial bloom identification results were determined using a multi-index consensus method, the formula for which is: ； ； ； ； Where i represents the pixel number in the image. , and This represents the binarization result of pixel i under NDVI, RVI and DVI indices, where 1 represents cyanobacterial bloom and 0 represents non-cyanobacterial bloom; This represents the final result identified by the multi-index consensus method for pixel i, where 1 represents cyanobacterial bloom and 0 represents non-cyanobacterial bloom.
4. The intraday dynamic prediction method for cyanobacterial blooms based on meteorological-water quality synergistic driving as described in claim 3, characterized in that, The formula for calculating the actual coverage area of the cyanobacterial bloom is as follows: ； ； Where n is the total number of pixels affected by cyanobacterial blooms, and i is the pixel number affected by cyanobacterial blooms. Let i be the area of the i-th cyanobacterial bloom pixel. This represents the actual area covered by cyanobacterial blooms. Let i be the cyanobacterial bloom coverage index for the i-th pixel. This represents the normalized vegetation index of the i-th pixel. and The NDVI values are -0.2 and 0.81 for clean water bodies and cyanobacterial blooms, respectively.
5. The intraday dynamic prediction method for cyanobacterial blooms based on meteorological-water quality synergistic driving as described in claim 4, characterized in that, The formula for calculating the mean absolute error is: ； In the formula, The mean absolute error, Represents the actual observed value. This represents the predicted value.
6. The intraday dynamic prediction method for cyanobacterial blooms based on meteorological-water quality synergistic driving as described in claim 5, characterized in that, The optimization of a BP neural network model includes the following steps: a) Normalize the hourly data of actual coverage area of cyanobacterial blooms and corresponding meteorological factors and water quality parameters as sample data to obtain normalized sample data. b) Input the normalized sample data into the BP neural network, randomly set the initial parameters and threshold, and set the number of hidden layer neurons to 5 to 30, and set the number of iterations to 1000 for each neuron's data volume; c) Train the BP neural network, output the network prediction results and determine whether the maximum number of iterations has been reached. If not, return to step b). If so, increase the number of hidden layer neurons in turn with a step size of 1, and determine whether the number of neurons exceeds the preset range. If so, complete the training. If not, return to step b). 26,000 network models with different numbers of hidden layer neurons and initial parameter settings are obtained through iterative training.
7. The intraday dynamic prediction method for cyanobacterial blooms based on meteorological-water quality synergistic driving as described in claim 6, characterized in that, The efficiency coefficient for calculating the estimated coverage area of cyanobacterial blooms versus the actual coverage area of cyanobacterial blooms includes the following steps: Based on different network models, the estimated coverage area of cyanobacterial blooms predicted by each model is output. Calculate the average absolute error between the predicted cyanobacterial bloom coverage area and the actual cyanobacterial bloom coverage area of each network model. The network model with the smallest mean absolute error was selected as the prediction model, and the efficiency coefficient between the predicted actual coverage area of cyanobacterial blooms and the actual coverage area of real cyanobacterial blooms was calculated.
8. The intraday dynamic prediction method for cyanobacterial blooms based on meteorological-water quality synergistic driving as described in claim 7, characterized in that, The formula for calculating the efficiency coefficient is: ； In the formula, E represents the efficiency coefficient. Represents the actual observed value. Indicates the predicted value. This represents the average of the actual observed values, where n is the total number of samples.