Blue-green algae outbreak prediction method and system based on low-altitude multispectral remote sensing

Abstract

The invention provides a blue-green algae outbreak prediction method and system based on low-altitude multispectral remote sensing, and belongs to the technical field of environmental monitoring, and the method comprises the steps: obtaining low-altitude multispectral remote sensing monitoring data; constructing a blue-green algae core parameter inversion model, and obtaining remote sensing characteristic parameters through the blue-green algae core parameter inversion model; performing space-time matching on the meteorological data and the low-altitude multispectral remote sensing monitoring data, and obtaining a meteorological comprehensive influence index according to a matching result; spatial interpolation of hydrological data is realized through the hydrological dynamic model, and hydrological dynamic influence indexes are constructed according to an interpolation result; introducing a dynamic period length, and combining the remote sensing characteristic parameter and the dynamic period length to obtain a blue-green algae growth characteristic parameter; and fusing the remote sensing characteristic parameters, the meteorological comprehensive influence index, the hydrodynamic influence index and the blue-green algae growth characteristic parameters to obtain a blue-green algae outbreak prediction result. According to the invention, accurate estimation of blue-green algae biomass, deep coupling of multi-source data and dynamic period adaptation are realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of environmental monitoring, and particularly relates to a cyanobacteria outbreak prediction method and system based on low-altitude multispectral remote sensing. BACKGROUND

[0002] Cyanobacteria outbreak can lead to a decrease in water dissolved oxygen and the death of aquatic organisms, and the algal toxins produced can also threaten human health, so it is of great significance to achieve early and accurate prediction of cyanobacteria outbreak. The current mainstream cyanobacteria prediction method mainly relies on satellite remote sensing and statistical analysis of multi-source data, but has the following technical defects:

[0003] Insufficient accuracy of remote sensing data processing: satellite remote sensing has low spatial resolution, making it difficult to capture the early small-scale aggregation characteristics of cyanobacteria; and traditional atmospheric correction uses fixed aerosol optical thickness and ignores the dynamic changes of near-surface water vapor, resulting in large errors in the inversion of surface reflectivity, which affects the accuracy of subsequent cyanobacteria parameter inversion.

[0004] One-sidedness of cyanobacteria core parameter inversion: existing methods mostly rely on chlorophyll a (Chla) concentration to invert cyanobacteria biomass, without considering the unique phycocyanin of cyanobacteria (an important component of cyanobacteria biomass), and the inversion coefficient uses a fixed value, which cannot adapt to seasonal, water temperature, and nutrient salt changes, resulting in a biomass estimation error of >20%.

[0005] Poor coordination of multi-source data: the spatio-temporal matching of meteorological data (temperature, light), hydrological data (flow rate, water level), and remote sensing data only uses simple interpolation, without considering the physiological characteristics of cyanobacteria (such as the influence of wind speed on cyanobacteria aggregation) and the continuity of hydrodynamic characteristics, resulting in low effectiveness of the features after multi-source data fusion.

[0006] Limited prediction lead time and accuracy: traditional time series analysis (such as fixed period STL decomposition) cannot capture the short-term fluctuations and dynamic periods of cyanobacteria growth, and the prediction model (such as ordinary LSTM) does not focus on key features before the outbreak (such as abnormal rise in biomass + suitable weather), resulting in a short prediction lead time and low accuracy of outbreak probability.

[0007] In summary, existing technologies cannot meet the prediction needs of "earlier and more accurate" cyanobacteria outbreak, and there is an urgent need for a technical solution that takes into account the accuracy of remote sensing data, the coordination of multi-source data, and the timeliness of prediction. SUMMARY

[0008] To solve the above problems, the application provides a cyanobacteria outbreak prediction method and system based on low-altitude multispectral remote sensing, and the specific technical solution is as follows:

[0009] A cyanobacteria outbreak prediction method based on low-altitude multispectral remote sensing, comprising the following steps:

[0010] Obtain low-altitude multispectral remote sensing monitoring data;

[0011] The low-altitude multi-spectral remote sensing monitoring data is used to construct a cyanobacteria core parameter inversion model, and a remote sensing characteristic parameter is obtained through the cyanobacteria core parameter inversion model; meteorological data and the low-altitude multi-spectral remote sensing monitoring data are spatio-temporally matched, and a meteorological comprehensive influence index is obtained according to a matching result;

[0012] A hydrological dynamic model is used to realize spatial interpolation of hydrological data, and a hydrological dynamic influence index is constructed according to an interpolation result;

[0013] A dynamic cycle length is introduced, and a cyanobacteria growth characteristic parameter is obtained in combination with the remote sensing characteristic parameter and the dynamic cycle length;

[0014] The remote sensing characteristic parameter, the meteorological comprehensive influence index, the hydrological dynamic influence index and the cyanobacteria growth characteristic parameter are fused to obtain a cyanobacteria outbreak prediction result.

[0015] A cyanobacteria outbreak prediction system based on low-altitude multi-spectral remote sensing includes the following modules:

[0016] A data acquisition module acquires low-altitude multi-spectral remote sensing monitoring data;

[0017] A remote sensing characteristic parameter acquisition module uses the low-altitude multi-spectral remote sensing monitoring data to construct a cyanobacteria core parameter inversion model, and a remote sensing characteristic parameter is obtained through the cyanobacteria core parameter inversion model;

[0018] A meteorological comprehensive influence index acquisition module spatio-temporally matches meteorological data and the low-altitude multi-spectral remote sensing monitoring data, and a meteorological comprehensive influence index is obtained according to a matching result;

[0019] A hydrological dynamic influence index acquisition module realizes spatial interpolation of hydrological data through a hydrological dynamic model, and a hydrological dynamic influence index is constructed according to an interpolation result;

[0020] A cyanobacteria growth characteristic parameter acquisition module introduces a dynamic cycle length, and a cyanobacteria growth characteristic parameter is obtained in combination with the remote sensing characteristic parameter and the dynamic cycle length;

[0021] A cyanobacteria outbreak result prediction module fuses the remote sensing characteristic parameter, the meteorological comprehensive influence index, the hydrological dynamic influence index and the cyanobacteria growth characteristic parameter to obtain a cyanobacteria outbreak prediction result.

[0022] An electronic device includes one or more processors; a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method.

[0023] A computer-readable storage medium having stored thereon executable instructions that, as a result of being executed by a processor, enable the processor to implement the described method.

[0024] The present application has the following beneficial effects:

[0025] The present application introduces a dynamic water vapor correction coefficient to construct a dynamic atmospheric correction model, which can specifically eliminate the interference of low-altitude near-surface water vapor dynamic changes on remote sensing data (such as reflectivity deviation caused by water vapor pressure difference at different times), avoid the precision limitations of traditional fixed parameter correction, make the corrected surface reflectivity more in line with the actual situation, and lay a precise data foundation for subsequent cyanobacteria parameter inversion; by introducing the noise reduction model of the cyanobacteria spectral attention mechanism, the spectral information of the cyanobacteria characteristic bands (680nm chlorophyll peak, 700nm red edge band) can be strengthened during noise reduction, the traditional noise reduction model can avoid blurring the subtle spectral differences between cyanobacteria and water, effectively retain key features such as cyanobacteria aggregation area and small patches, and improve the accuracy of subsequent cyanobacteria identification and parameter inversion.

[0026] The present application constructs a chlorophyll a and phycocyanin dual inversion sub-model, identifies cyanobacteria through phycocyanin specificity, solves the problem that traditional chlorophyll a cannot distinguish cyanobacteria from other algae, avoids misjudgment, dynamically adjusts the inversion coefficient with historical remote sensing data to replace the fixed coefficient, adapts to environmental changes such as seasons, water temperature and water quality, and reduces inversion deviation; finally, through the dual model corrected by the dynamic coefficient, the remote sensing characteristic parameters such as chlorophyll a, phycocyanin and cyanobacteria biomass are accurately obtained, providing high-quality and high-specificity core data support for subsequent multi-source fusion and prediction.

[0027] The present application introduces meteorological factor spatial heterogeneity weight (combining wind speed to dynamically adjust interpolation weight), avoids the problem that simple interpolation ignores the spatial difference of meteorological elements, makes the spatial interpolation result more in line with the low-altitude remote sensing pixel scale demand, and ensures the timeliness of the data through time matching; based on the accurate meteorological data after time and space matching, an integrated influence index model is constructed combining the physiological characteristics of cyanobacteria, the "promotion-inhibition" effect of temperature, illumination and other factors on the growth of cyanobacteria is quantified, the limitations of traditional indexes with weak correlation with cyanobacteria growth are avoided, and high-correlation meteorological feature support is provided for subsequent multi-source fusion and prediction.

[0028] The present application corrects Kriging interpolation with hydrodynamic model flow field data, dynamically adjusts the flow velocity interpolation result through flow difference, ensures the spatial continuity and dynamic consistency of hydrological data (such as flow velocity), avoids errors caused by ignoring the flow rule in simple interpolation, and is in line with the actual motion state of the water body; based on the accurate interpolation result, a hydrodynamic influence index model is constructed, the "promotion-inhibition" effect of flow velocity, water level and other factors on the growth of cyanobacteria (such as low flow velocity conducive to the residence of cyanobacteria) is quantified, the problem of weak correlation between traditional indexes and cyanobacteria growth is solved, and high-precision and high-correlation hydrological feature support is provided for subsequent multi-source fusion and prediction.

[0029] The application introduces a dynamic cycle length instead of a fixed cycle, dynamically determines a blue-green algae growth cycle (such as a quarter or a year) based on an autocorrelation function, adapts to the cycle change of blue-green algae growth in different seasons and environments, and avoids cycle misjudgment; based on the decomposed trend item, cycle item and residual item, the growth characteristic parameters such as trend change rate and periodicity intensity are accurately extracted, and the long-term growth trend and short-term fluctuation (such as temporary decline after rainstorm) of blue-green algae are clearly captured; finally, the dynamic characteristics reflecting the growth law of blue-green algae are provided for subsequent multi-source fusion and prediction, which helps to identify the precursor of outbreak in advance and prolong the prediction lead time.

[0030] The application firstly correlates blue-green algae physiological parameters and meteorological suitability, and then couples hydrodynamic and growth law, so as to avoid information redundancy caused by feature stacking; a prediction model is constructed based on classification features, rather than directly inputting original multi-source data, so as to accurately capture the synergistic effect of different dimensions on blue-green algae outbreak (such as high-risk combination of meteorological suitability + hydrological advantage + growth rise); finally, the model outputs probability, time and driving factor, which not only improves the prediction accuracy, but also clearly identifies the key inducement of outbreak, provides a scientific basis for targeted prevention and control, and solves the problem of traditional prediction "knowing the result but not knowing the reason".

[0031] The application is designed in a closed loop of "data acquisition-parameter inversion-multi-source fusion-prediction output", takes low-altitude multi-spectral remote sensing as the core, constructs a blue-green algae core parameter inversion model, accurately obtains chlorophyll a, phycocyanin and other parameters, solves the problem of one-sidedness of traditional single index inversion, and provides a reliable basis for prediction; meteorological, hydrological and remote sensing data are deeply coupled, time and space matching ensures the timeliness of meteorological influence index, and the interpolation of hydrodynamic model improves the spatial continuity of hydrological index, breaking the limitation of loose combination of multi-source data; the dynamic cycle length is introduced to capture the growth law of blue-green algae, multi-feature fusion is introduced to strengthen the prediction relevance, earlier and more accurate outbreak prediction is realized, scientific guidance is provided for lake blue-green algae prevention and control, and ecological and health risks are reduced. BRIEF DESCRIPTION OF DRAWINGS

[0032] Figure 1 A blue-green algae outbreak prediction method flowchart based on low-altitude multi-spectral remote sensing is provided for the embodiments of the application;

[0033] Figure 2 A blue-green algae outbreak prediction system framework diagram based on low-altitude multi-spectral remote sensing is provided for the embodiments of the application. DETAILED DESCRIPTION

[0034] Specific embodiments of the present application will now be described in detail with reference to the following figures. A person of ordinary skill in the art will immediately appreciate that the embodiments described herein are merely for illustration and should not be taken as limiting the application. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. It will be apparent, however, to one skilled in the art that the present application can be practiced without the specific details. In other instances, well-known circuits, software or methods have not been described in detail in order to avoid obscuring the present application. It will be appreciated that the various units and / or features can be implemented in hardware, software or a combination thereof, and can be implemented as part of one or more computer programs.

[0035] Reference throughout this specification to "an embodiment", "embodiments", "one example", or "an example" means that a particular feature, structure, or characteristic described in connection with the embodiment or example is included in at least one embodiment of the application. Therefore, appearances of the phrases "in one embodiment", "in embodiments", "one example" or "an example" in various places throughout this specification are not necessarily referring to the same embodiment or example. Furthermore, the particular features, structures, or characteristics can be combined in any suitable

[0036] Referring to Figure 1 The present application provides a cyanobacteria bloom prediction method based on low-altitude multispectral remote sensing, as shown in Figure 1 The method includes the following steps in one embodiment:

[0037] S1, obtaining low-altitude multispectral remote sensing monitoring data.

[0038] Low-altitude unmanned aerial vehicles or small aircraft equipped with high-performance multispectral sensors are used to obtain low-altitude multispectral remote sensing monitoring data. The data collection covers the 400-900 nm visible light-near infrared band, including key bands such as blue light (450-520 nm), green light (520-600 nm), red light (630-690 nm), and near infrared (760-900 nm). The spatial resolution is set to 1-5 m, which can clearly capture the changes in water details. The time resolution is 1 day, which realizes high-frequency dynamic monitoring.

[0039] Further, auxiliary data is obtained through a multi-source data acquisition system: real-time monitoring of water vapor pressure (unit: hPa) using a low-altitude weather station, which affects atmospheric humidity and spectral transmission characteristics; ground-based sun photometer is used to measure the ground aerosol optical depth, which reflects the interference of atmospheric turbidity on remote sensing signals; the water body turbidity (unit: NTU) is collected by a portable water quality detector, which quantifies the scattering and absorption of suspended particles in water on the spectrum; based on high-precision ephemeris data and imaging time information, the sun elevation angle (unit: °) at the remote sensing imaging time is accurately calculated, which affects the intensity and direction of ground object reflected spectrum.

[0040] The traditional 6S model, i.e., a secondary simulation model of satellite signals under sunlight spectrum and a dynamic atmospheric correction submodule, does not consider the influence of water body turbidity on near-surface radiation scattering (high turbidity water body is prone to additional scattering, which interferes with the calculation of ground reflectivity), and does not adjust the atmospheric path radiation according to the daily variation of the solar elevation angle (such as high solar elevation angle at noon and low solar elevation angle in the morning / evening). The application introduces a dynamic water vapor correction coefficient to construct a dynamic atmospheric correction model, and uses the dynamic atmospheric correction model to correct the low-altitude multispectral remote sensing monitoring data.

[0041] Specifically, the dynamic atmospheric correction model satisfies the following formula:

[0042] ;

[0043] Wherein, represents the ground reflectivity after dynamic correction, represents the wavelength, represents the top space radiation received by the sensor, represents the atmospheric path radiation, represents the solar elevation angle at the remote sensing imaging time, represents the average reflectivity of adjacent pixels, which is the average value of the reflectivity of 3x3 pixels around the target pixel, represents the sky diffuse reflection radiation, represents the turbidity correction coefficient, represents the atmospheric spherical reflection coefficient, represents the variation coefficient of aerosol optical depth with wavelength, represents the real-time aerosol optical depth, represents the dynamic water vapor correction coefficient, which is the water vapor pressure monitored by the low-altitude meteorological station in real time calculated, .

[0044] Further, the turbidity correction coefficient reflects the enhancement effect of water body turbidity on sky diffuse reflection radiation. When the concentration of suspended particles in the water body increases, the Mie scattering and Rayleigh scattering of particles intensify, so that the intensity of sky diffuse reflection radiation scattered by the water body increases significantly. This coefficient quantifies this enhancement relationship and provides a key parameter for the atmospheric correction model to eliminate the influence of turbidity interference on cyanobacterial spectral feature extraction.

[0045] The calculation is based on synchronous water body turbidity monitoring data (obtained by a water quality monitoring station, unit: NTU), and a segmented function is used to dynamically determine: when the turbidity is <10 NTU, the water body is in a low turbidity state, the particle scattering effect is weak, and a linear regression equation is used to calculate, wherein The turbidity value is measured; when the turbidity is greater than 10 and less than 50 NTU, the turbidity is in the medium-high interval, the particle scattering is enhanced and presents a nonlinear change, and the turbidity is fitted by a quadratic polynomial Dynamic calibration is realized; when the turbidity is greater than 50 NTU, the water body is highly turbid, and an empirical constant is used Fast response to extreme turbidity conditions ensures model robustness. The parameters of each segmented function are based on the cross-validation optimization of historical monitoring data and multispectral remote sensing inversion results, and can be further optimized and adjusted in other embodiments.

[0046] A cyanobacterial spectral attention mechanism is introduced to construct a denoising model, and the low-altitude multispectral remote sensing monitoring data is denoised according to the denoising model.

[0047] The denoising model takes a symmetric encoder-decoder deep learning network as a basic model, adds a cyanobacterial spectral attention mechanism to the last two layers of the decoder, gives higher weights to cyanobacterial characteristic bands (680 nm, 700 nm) (such as generating a band weight map through a Sigmoid activation function, so that the 680 nm band weight is 30%-50% higher than that of non-characteristic bands), avoids blurring the spectral differences between cyanobacteria and water bodies during denoising, and uses a weighted sum of spectral angle matching loss and mean square error as the loss function. The specific formula is as follows:

[0048] ;

[0049] Among them, SINR680 represents the signal-to-noise ratio of the 680 nm band, SINR700 represents the signal-to-noise ratio of the 700 nm band, Spectral angle matching function, by calculating the cosine value of the angle between the corrected image and the true value in the spectral dimension, quantifying the similarity of the spectral curves of the two, the smaller the angle, the higher the spectral similarity, which can effectively reflect the matching of the cyanobacterial spectral characteristics, High-resolution reference image, i.e. true value, as the benchmark data for model training, Mean square error. The band with higher signal-to-noise ratio contains more effective information of cyanobacteria and is less affected by noise, so it has a greater contribution to cyanobacterial identification, and therefore is given a higher weight, so that the loss function can dynamically adjust the optimization direction according to the data characteristics, further improving the accuracy and robustness of the cyanobacterial prediction model.

[0050] S2, using the low-altitude multispectral remote sensing monitoring data to construct a cyanobacterial core parameter inversion model, obtaining a remote sensing characteristic parameter through the cyanobacterial core parameter inversion model, including the following steps:

[0051] S21, construct a chlorophyll a concentration inversion sub-model and a phycocyanin concentration inversion sub-model, and dynamically adjust inversion coefficients of the chlorophyll a concentration inversion sub-model and the phycocyanin concentration inversion sub-model through historical low-altitude multi-spectral remote sensing monitoring data.

[0052] The chlorophyll a concentration inversion sub-model satisfies:

[0053] ;

[0054] wherein, Chl a represents the chlorophyll a concentration, a represents the inversion coefficient, R750, R680 and R440 represent corrected reflectances at 750 nm, 680 nm and 440 nm wave bands, and are used to compensate for the influence of non-uniform distribution caused by algae aggregation on the inversion result, Fv represents the algae coverage.

[0055] The phycocyanin concentration inversion sub-model satisfies:

[0056] ;

[0057] wherein, PC represents the phycocyanin concentration, a represents the inversion coefficient, R620 and R560 represent corrected reflectances at 620 nm and 560 nm wave bands.

[0058] In order to break through the limitation of traditional fixed coefficients, let the inversion coefficients be dynamically adjusted according to water temperature, pH and total phosphorus, adapt to environmental changes, solve the problem of large inversion error in non-suitable environment, and make the inversion more scientific by fitting the physiological mechanism of blue-green algae; at the same time, in order to reduce the dependence on measured data and avoid a large number of water sampling for fitting every month, only routine environmental parameters are needed, the cost is reduced, the coefficients are updated every day, the timeliness and precision are considered, and a foundation is laid for subsequent accurate inversion of blue-green algae parameters. The inversion coefficients satisfy:

[0059] ;

[0060] ;

[0061] ;

[0062] ;

[0063] wherein, Tw represents the water temperature, TP represents the total phosphorus content, pH represents the pH value.

[0064] The construction of the coefficient formula is based on the physiological characteristics of cyanobacteria (such as the synthesis of chlorophyll a depending on water temperature and phosphorus, and the stability of phycocyanin depending on pH), rather than simply fitting data, making the inversion process more in line with the physical-biological mechanism of cyanobacterial growth. For example, the coefficient of a positive change with TP (0.05) corresponds to the ecological theory that "phosphorus is a limiting factor for the growth of cyanobacteria", ensuring that the coefficient change has a clear scientific basis rather than being empirically adjusted.

[0065] S22, obtaining remote sensing characteristic parameters according to the cyanobacteria core parameter inversion model after dynamically adjusting the inversion coefficient.

[0066] Specifically, the phycocyanin-chlorophyll a ratio coefficient is introduced to calculate the cyanobacterial biomass, and the weights of the two are dynamically determined based on the measured data, and the formula is as follows:

[0067] ;

[0068] Wherein, represents the cyanobacterial biomass, represents the phycocyanin-chlorophyll a ratio coefficient, reflecting the actual ratio of chlorophyll a to phycocyanin in cyanobacterial cells, , and the coefficient 8 is based on the empirical relationship between the dry weight of cyanobacteria and the pigment concentration (cyanobacterial dry weight ≈ 8 × (chlorophyll a + phycocyanin) concentration).

[0069] Further, the remote sensing characteristic parameters include the time series of chlorophyll a concentration, the time series of phycocyanin concentration, the cyanobacterial biomass, the time series of phycocyanin-chlorophyll a ratio, the cyanobacterial coverage, the cyanobacterial aggregation index and the cyanobacterial specificity index.

[0070] The cyanobacterial aggregation index is the ratio of the corrected reflectance at 680 nm and 560 nm; the cyanobacterial specificity index is the ratio of phycocyanin concentration to chlorophyll a concentration, which is used to distinguish cyanobacteria from other algae: the cyanobacterial specificity index of cyanobacteria is about 0.8-1.2, and the non-cyanobacteria is less than 0.3, solving the problem that Chla alone cannot distinguish between algae species.

[0071] S3, spatiotemporal matching of meteorological data and the low-altitude multispectral remote sensing monitoring data, obtaining a meteorological comprehensive influence index according to the matching result. Including the following steps:

[0072] S31, spatial interpolation and time matching using meteorological factor spatial heterogeneity weight.

[0073] Spatiotemporal matching of meteorological data and remote sensing data (spatial interpolation to remote sensing image pixel scale, time matching to remote sensing imaging time ±1h). Spatial interpolation adopts inverse distance weighted interpolation, and introduces meteorological factor spatial heterogeneity weight. The greater the wind speed, the weaker the spatial correlation, and the faster the interpolation weight decays, and the meteorological factor spatial heterogeneity weight satisfies:

[0074] ;

[0075] wherein, represents the weight of the i-th weather station, represents the distance from the pixel to the i-th station, represents the interpolated wind speed of the pixel. When the wind speed is large, the spatial variation of meteorological elements is fast, and the weight of the nearby station is higher.

[0076] Further, the time matching adopts linear interpolation, and if there is no meteorological data at the time of remote sensing imaging, the data of 1 hour before and after is interpolated.

[0077] S32, constructing a meteorological comprehensive influence index model, based on the spatio-temporal matching result and the meteorological comprehensive influence index model, obtaining the meteorological comprehensive influence index.

[0078] The meteorological comprehensive influence index model satisfies the following formula:

[0079] ;

[0080] wherein, represents the meteorological comprehensive influence index, represents the weight of each meteorological factor, represents the temperature influence function, represents the light influence function, represents the wind speed influence function, represents the wind direction influence function.

[0081] The weight of the meteorological factor is dynamically determined based on the Pearson correlation coefficient. The stronger the correlation with the growth of blue-green algae, the greater the weight, which satisfies: wherein represents the correlation coefficient of the meteorological factor x and the blue-green algae biomass.

[0082] Further, the temperature influence function adopts a piecewise function form to quantitatively characterize the influence of temperature on the growth of blue-green algae, reflecting the suitable temperature interval characteristics of blue-green algae growth. When less than 25℃, the growth of blue-green algae is limited by temperature, and the function shows a linear increasing trend, indicating that in the low temperature range, with the increase of temperature, the suitability of blue-green algae growth gradually improves. When the temperature is greater than 25℃ and less than 30℃, it is the best temperature range for the growth of blue-green algae, at which the suitability of blue-green algae growth reaches the highest and remains stable. When greater than 30℃, the high temperature begins to inhibit the growth of blue-green algae, and the function shows a linear decrease, meaning that with the continuous increase of temperature, the suitability of blue-green algae growth gradually decreases. The function value range is 0-1, 0 indicates that the temperature condition is completely unsuitable for the growth of blue-green algae, and 1 indicates that the temperature condition is most favorable for the growth of blue-green algae.

[0083] The light intensity effect function quantifies the limiting effect of light intensity on cyanobacterial growth, providing key parameters for cyanobacterial bloom prediction models. Specifically, when the light intensity is below 400 W / m², the effect is limited. 2 When the light intensity is insufficient to meet the photosynthetic needs of cyanobacteria, the light influence function value is 0, indicating that light becomes a strong limiting factor for cyanobacterial growth; when the light intensity is 400 W / m², the light influence function value is 0. 2 Up to 600W / m 2 Between these values, the function value increases linearly with increasing light intensity, indicating that suitable light conditions can promote cyanobacteria growth; however, when the light intensity exceeds 600 W / m², the function value decreases. 2 Excessive light intensity can damage the cell structure and photosynthetic mechanism of cyanobacteria. The light effect function value is 1, indicating that light intensity once again becomes a key factor limiting cyanobacterial growth. The physical significance of this function lies in its precise characterization of the nonlinear relationship between light intensity and cyanobacterial growth, specifically below 400 W / m². 2 or higher than 600W / m 2 At the same time, light intensity limits the growth of cyanobacteria, providing a quantitative basis for predicting cyanobacterial blooms based on light conditions.

[0084] The wind speed influence function satisfies: The function value ranges from 0 to 1. Based on the migration and diffusion characteristics of cyanobacteria in water, its physical meaning is as follows: when wind speed increases, surface disturbance intensifies, prompting cyanobacteria communities to spread over a wider area, reducing the degree of cyanobacteria aggregation in local areas, thus weakening the impact of cyanobacteria aggregation on environmental monitoring indicators and the ecosystem. Conversely, when wind speed is low, cyanobacteria tend to aggregate in local areas, leading to an increase in cyanobacteria concentration in those areas and having a more significant impact on the surrounding water environment and ecological balance. This function quantifies the degree of influence of wind speed on the spatial distribution of cyanobacteria, providing key environmental impact parameters for cyanobacteria outbreak prediction models.

[0085] The wind direction influence function quantifies the influence of wind direction on the aggregation of blue-green algae. In actual lake ecological environments, the direction of wind determines the migration direction of blue-green algae, which in turn affects the likelihood and scale of blue-green algae blooms. When the wind direction points to the center of the lake, blue-green algae will accelerate towards the lake center under the influence of wind, leading to a large number of blue-green algae breeding and growing in a relatively concentrated space, significantly increasing the risk of a bloom. In this case, the wind direction influence function is defined as 1, indicating that the wind direction is the most favorable condition for promoting a blue-green algae bloom. When the wind direction points to the shore, blue-green algae are driven towards the shore by the wind. Since the shore area is relatively open and the water flow is enhanced, it is not conducive to the large-scale aggregation of blue-green algae, reducing the probability of a blue-green algae bloom. Therefore, the wind direction influence function is set to 0.5, indicating that the wind direction has a weak promoting effect on a blue-green algae bloom. The physical meaning of this function is to accurately depict the internal relationship between wind direction and the aggregation direction of blue-green algae by numerical expression, and to clearly indicate that the wind direction towards the center is more likely to form a bloom, providing a key quantitative indicator for the wind direction based on low-altitude multispectral remote sensing of blue-green algae bloom prediction models.

[0086] Further, to reflect the inhibitory effect of extreme weather on blue-green algae, a meteorological anomaly factor is constructed. When the temperature is greater than 35℃ or V>5m / s, the meteorological anomaly factor is 1, otherwise it is 0.

[0087] S4, spatial interpolation of hydrological data is realized by a hydrological dynamic model, and a hydrological dynamic influence index is constructed according to the interpolation results, including the following steps:

[0088] S41, the kriging interpolation is corrected by the flow field data output by the hydrological dynamic model, and the spatial interpolation of hydrological data is realized by using the corrected results.

[0089] Specifically, the kriging interpolation is corrected in combination with the flow field data output by the MIKE21 hydrodynamic model, and the following conditions are met:

[0090] ;

[0091] wherein, V(x,y) represents the flow velocity of the pixel (x,y), K(x,y) represents the kriging interpolation, D(x,y) represents the flow direction of the pixel (x,y), D(x,y) represents the flow direction of the adjacent hydrological station. The greater the flow direction difference, the weaker the flow field continuity, and the interpolation result needs to be adjusted downward to ensure the dynamic consistency of the flow velocity and the flow direction.

[0092] S42, a hydrological dynamic influence index model is constructed, and a hydrological dynamic influence index is obtained by combining the hydrological dynamic influence index model and the spatial interpolation results.

[0093] In the embodiment, the hydrological dynamic influence index model satisfies the following formula:

[0094] ;

[0095] wherein, represents a hydrological dynamic influence index, represents a weight of each hydrological factor, satisfying: , represents a correlation coefficient of the hydrological factor y and the cyanobacterial biomass, represents a flow rate influence function, represents a water level influence function, represents a flow direction influence function.

[0096] The flow rate influence function satisfies: When the flow rate is less than 0.2 m / s, , it is conducive to the retention of cyanobacteria; when the flow rate is greater than 0.5 m / s, , it inhibits aggregation.

[0097] The water level influence function satisfies: wherein, represents a water level, represents a historical highest water level of the lake, represents a water level change rate. The higher the water level, the larger the cyanobacterial diffusion space; and a too rapid water level drop inhibits the growth of cyanobacteria.

[0098] The flow direction influence function is that, if the flow direction is consistent with the cyanobacterial aggregation area (identified by remote sensing), the flow direction influence function value is 1, otherwise it is 0.6. When the flow direction is consistent with the aggregation direction, it accelerates the convergence of cyanobacteria.

[0099] S5, introducing a dynamic cycle length, obtaining a cyanobacterial growth characteristic parameter in combination with the remote sensing characteristic parameter and the dynamic cycle length, including the following steps:

[0100] S51, introducing a dynamic cycle length to perform time series decomposition on the remote sensing characteristic parameter.

[0101] Specifically, for time series sample data, the lag order range is set to 30-400 days (covering the possible quarterly period (30-90 days), semi-annual period (180 days), and annual period (365 days) of cyanobacteria); then, the autocorrelation function value of each lag order is calculated, and the lag order that first exceeds 0.6 and is the largest is selected as the initial dynamic cycle length P.

[0102] Finally, the autocorrelation function values of the cyanobacterial biomass time series and the phycocyanin concentration time series are calculated, and if the difference between the dynamic cycle lengths P of the two is ≤15 days, the average of the two is taken as the final dynamic cycle length P; if the difference is >15 days, the sample subset in the cyanobacterial growth stage (such as the summer outbreak period) is used to recalculate to ensure that the dynamic cycle length P meets the cyanobacterial physiological cycle.

[0103] Furthermore, the remote sensing feature parameters are decomposed into time series to satisfy:

[0104] ;

[0105] ;

[0106] ;

[0107] in, It indicates a long-term growth trend, used to depict the overall direction of change in cyanobacteria populations over a longer timescale, such as a stable growth or decline trend resulting from factors like seasonal changes and environmental changes. Indicates seasonal cycles, express Seasonal cycles of time This indicates short-term fluctuations in the residual. Indicates the trend adjustment factor. The trend update step size is dynamically adjusted based on the direction and magnitude of the trend change. When the cyanobacteria growth trend accelerates, the adjustment magnitude is increased, and when the trend slows down, the adjustment magnitude is decreased, thereby achieving adaptive fitting to complex growth trends. This represents the rate of change of the long-term trend term of cyanobacterial biomass. Represents the periodic intensity factor. , It represents the standard deviation of the sequence. By comparing the standard deviation of the periodic term with that of the original sequence, the amplitude of periodic fluctuations is dynamically adjusted to ensure accurate matching of periodic changes under different environments.

[0108] S52. Based on the decomposition results, extract the characteristic parameters of cyanobacteria growth.

[0109] The "Define Trend Type" label is based on the trend change rate; for example, a trend change rate greater than 0.5 indicates a "rapid rise." The "Cycle Stability" label is based on the cycle strength. ,For example, "Highly significant periodicity"; based on residual abnormal factors Establish "anomaly-recovery" identification rules, if And the following 3 days and If it is determined to be "rapid recovery after an anomaly" (such as the re-aggregation of cyanobacteria after a rainstorm, which requires close monitoring); if And the following 7 days It was determined to be "post-abnormal suppression" (e.g., cyanobacteria are difficult to recover after strong winds, reducing the risk of an outbreak). This represents the rate of change of the long-term trend term of cyanobacterial biomass at time t.

[0110] Further, the cyanobacteria growth characteristic parameter further comprises a trend-cycle coordination characteristic, satisfying: ; represents the trend-cycle coordination characteristic at time t; if it is positive and greater than 0.3, it represents “rising trend + significant period”, and the outbreak risk is high; and if it is negative, it represents “falling trend”, and the risk is low.

[0111] S6, fusing the remote sensing characteristic parameter, the meteorological comprehensive influence index, the hydrological dynamic influence index and the cyanobacteria growth characteristic parameter, obtaining a cyanobacteria outbreak prediction result, comprising the following steps:

[0112] S61, combining the remote sensing characteristic parameter and the meteorological comprehensive influence index, obtaining a remote sensing-meteorological characteristic, satisfying:

[0113] ;

[0114] wherein, represents the remote sensing-meteorological characteristic, represents the weight coefficient of the i-th remote sensing characteristic parameter, , represents the characteristic importance (calculated by a random forest model, based on the contribution of the characteristic to the cyanobacteria outbreak label, taking a value of 0-1), represents the growth stage weight (0.3 for the rising period, meteorological dominant; 0.5 for the stable period, balanced; 0.7 for the outbreak period, remote sensing biomass dominant; and 0.4 for the falling period), represents the normalization factor, , ensuring that the weight sum is 1, represents the i-th remote sensing characteristic parameter, represents the weight coefficient of the j-th meteorological comprehensive influence index, , represents the j-th meteorological comprehensive influence index. By introducing the cyanobacteria growth stage attention weight, the meteorological characteristic weight is increased in the early outbreak period (meteorological dominant growth), and the remote sensing characteristic weight is increased in the outbreak period (biomass dominant diffusion).

[0115] Further, the remote sensing-meteorological characteristic comprises a cyanobacteria biomass-meteorological influence index, a chlorophyll a-air temperature coupling index, a cyanobacteria coverage-illumination coupling index, and an aggregation index-wind speed coupling index.

[0116] S62, combining the hydrological dynamic influence index and the cyanobacteria growth characteristic parameter, obtaining a hydrological-time sequence correlation characteristic.

[0117] The mutual information is used to calculate the nonlinear correlation degree, and then the correlation characteristic is constructed to reflect the influence of the hydrological dynamics on the trend and period of the cyanobacteria time sequence.

[0118] First, mutual information between the hydrodynamic impact index and the cyanobacterial growth characteristic parameters is calculated.

[0119] Specifically, the mutual information calculation satisfies:

[0120]

[0121] Among them, represents the mutual information of features X and Y (0-∞, the greater the value, the stronger the nonlinear correlation).

[0122] Then, according to the mutual information, the hydrodynamic-trend correlation feature, the water level change-trend rate correlation feature, and the flow velocity-periodicity correlation feature are calculated.

[0123] The hydrological-timing correlation feature includes the hydrodynamic-trend correlation feature, the hydrodynamic impact index at time t The correlation between the long-term growth trend at time t The higher the mutual information, the stronger the influence of hydrology on the trend, the greater the feature weight, and the hydrodynamic-trend correlation feature satisfies:

[0124] The water level change-trend rate correlation feature, the water level change rate at time t The correlation between reflects the influence of water level change on the trend speed, and the water level change-trend rate correlation feature satisfies:

[0125] The flow velocity-periodicity correlation feature, the flow velocity at time t The correlation between the periodicity intensity at time t The smaller the flow velocity, the more significant the period, the higher the mutual information, and the flow velocity-periodicity correlation feature satisfies:

[0126] In addition, an abnormal correlation factor is added, if , indicating that the hydrology is suitable + the period is significant + the trend is rising, the outbreak risk is high, otherwise 0.

[0127] S63, a cyanobacterial outbreak prediction model is constructed, and the remote sensing-weather feature, the hydrological-timing correlation feature, and the cyanobacterial outbreak prediction model are combined to obtain a cyanobacterial outbreak prediction result, including the following steps:

[0128] A Bi-LSTM model is used to construct a cyanobacterial outbreak prediction model, and the remote sensing-weather feature and the hydrological-timing correlation feature are combined to calculate the feature attention layer weight. ​​​​

[0129] Specifically, the characteristic attention layer weight satisfies:

[0130] ;

[0131] ;

[0132] wherein, denotes the attention layer weight of the feature on the tth day, denotes a hydrological-time series correlation feature, denotes a weight matrix, denotes a bias.

[0133] Then, the burst probability and time prediction are performed according to the feature attention layer weight, and a cyanobacterial burst prediction result is obtained.

[0134] Specifically, the burst probability prediction satisfies:

[0135] ;

[0136] wherein, denotes a future number of days, denotes a sigmoid function, , is determined as a high burst risk, is a medium risk, is a low risk. As an input sequence of the Bi-LSTM model, the core role is to extract deep features related to the burst from the bidirectional time series dependence, directly supporting long-acting and high-precision probability prediction.

[0137] Further, the burst time prediction satisfies: that is, the date with the highest burst probability is taken as the predicted burst time, and if all , , the output is "no burst risk in the future 7d".

[0138] Please refer to Figure 2 , in the embodiment, in order to efficiently execute the cyanobacterial burst prediction method provided by the application based on low-altitude multispectral remote sensing, the application further provides a cyanobacterial burst prediction system based on low-altitude multispectral remote sensing, comprising: input device, output device, processor, memory, the input device, output device, processor, memory are connected with each other, the memory contains program instructions, the program instructions are used for the steps of the cyanobacterial burst prediction method based on low-altitude multispectral remote sensing. The cyanobacterial burst prediction system based on low-altitude multispectral remote sensing provided by the application has compact structure and stable performance, can stably execute the cyanobacterial burst prediction method based on low-altitude multispectral remote sensing provided by the application, and further improves the overall applicability and practical application ability of the application.

Claims

1. A cyanobacterial bloom prediction method based on low-altitude multispectral remote sensing, characterized by, The method comprises the following steps: obtaining low-altitude multi-spectral remote sensing monitoring data; constructing a cyanobacteria core parameter inversion model using the low-altitude multi-spectral remote sensing monitoring data, and obtaining a remote sensing characteristic parameter through the cyanobacteria core parameter inversion model; spatiotemporally matching meteorological data and the low-altitude multi-spectral remote sensing monitoring data, and obtaining a meteorological comprehensive influence index according to the matching result; spatially interpolating hydrological data through a hydrological dynamic model, and constructing a hydrological dynamic influence index according to the interpolation result; introducing a dynamic cycle length, combining the remote sensing characteristic parameter and the dynamic cycle length to obtain a cyanobacteria growth characteristic parameter; fusing the remote sensing characteristic parameter, the meteorological comprehensive influence index, the hydrological dynamic influence index and the cyanobacteria growth characteristic parameter to obtain a cyanobacteria outbreak prediction result.

2. The cyanobacterial bloom prediction method based on low-altitude multispectral remote sensing according to claim 1, characterized in that, The method further comprises the following steps: introducing a dynamic water vapor correction coefficient to construct a dynamic atmospheric correction model, and correcting the low-altitude multi-spectral remote sensing monitoring data using the dynamic atmospheric correction model; introducing a cyanobacteria spectral attention mechanism to construct a noise reduction model, and denoising the low-altitude multi-spectral remote sensing monitoring data according to the noise reduction model.

3. The cyanobacterial bloom prediction method based on low-altitude multispectral remote sensing according to claim 1, characterized in that, The method of constructing a cyanobacteria core parameter inversion model using the low-altitude multi-spectral remote sensing monitoring data, and obtaining a remote sensing characteristic parameter through the cyanobacteria core parameter inversion model comprises the following steps: constructing a chlorophyll-a concentration inversion sub-model and a phycocyanin concentration inversion sub-model, and dynamically adjusting inversion coefficients of the chlorophyll-a concentration inversion sub-model and the phycocyllin concentration inversion sub-model through historical low-altitude multi-spectral remote sensing monitoring data; obtaining a remote sensing characteristic parameter according to the cyanobacteria core parameter inversion model after dynamic adjustment of the inversion coefficients.

4. The cyanobacterial bloom prediction method based on low-altitude multispectral remote sensing according to claim 1, characterized in that, The method of spatiotemporally matching meteorological data and the low-altitude multi-spectral remote sensing monitoring data, and obtaining a meteorological comprehensive influence index according to the matching result comprises the following steps: spatially interpolating and temporally matching using meteorological factor spatial heterogeneity weights; constructing a meteorological comprehensive influence index model, and obtaining a meteorological comprehensive influence index based on the spatiotemporal matching result and the meteorological comprehensive influence index model.

5. The cyanobacterial bloom prediction method based on low-altitude multispectral remote sensing according to claim 1, characterized in that, The method of spatially interpolating hydrological data through a hydrological dynamic model, and constructing a hydrological dynamic influence index according to the interpolation result comprises the following steps: correcting Kriging interpolation through flow field data output by the hydrological dynamic model, and spatially interpolating hydrological data using the correction result; constructing a hydrological dynamic influence index model, and obtaining a hydrological dynamic influence index in combination with the hydrological dynamic influence index model and the spatial interpolation result. 6.The cyanobacterial bloom prediction method based on low-altitude multispectral remote sensing of claim 1, wherein, The method of introducing a dynamic cycle length, combining the remote sensing characteristic parameter and the dynamic cycle length to obtain a cyanobacteria growth characteristic parameter comprises the following steps: introducing a dynamic cycle length to perform time series decomposition on the remote sensing characteristic parameter; extracting a cyanobacteria growth characteristic parameter according to the decomposition result.

7. The cyanobacterial bloom prediction method based on low-altitude multispectral remote sensing according to claim 1, characterized in that, The method of fusing the remote sensing characteristic parameter, the meteorological comprehensive influence index, the hydrological dynamic influence index and the cyanobacteria growth characteristic parameter to obtain a cyanobacteria outbreak prediction result comprises the following steps: combining the remote sensing characteristic parameter and the meteorological comprehensive influence index to obtain a remote sensing-meteorological characteristic; combining the hydrological dynamic influence index and the cyanobacteria growth characteristic parameter to obtain a hydrological-time series correlation characteristic; The blue algae outbreak prediction model is constructed, the remote sensing-weather feature, the hydrology-time sequence correlation feature and the blue algae outbreak prediction model are combined, and a blue algae outbreak prediction result is obtained.

8. The cyanobacterial bloom prediction method based on low-altitude multispectral remote sensing according to claim 7, characterized in that, The hydrology-time sequence correlation feature is obtained by combining the hydrology dynamic influence index and the blue algae growth feature parameter, and includes the following steps: Calculate the mutual information between the hydrology dynamic influence index and the blue algae growth feature parameter. According to the mutual information, the hydrology dynamic-trend correlation feature, the water level change-trend rate correlation feature and the flow rate-periodicity correlation feature are calculated.

9. The cyanobacterial bloom prediction method based on low-altitude multispectral remote sensing according to claim 7, characterized in that, The blue algae outbreak prediction model is constructed, the remote sensing-weather feature, the hydrology-time sequence correlation feature and the blue algae outbreak prediction model are combined, and a blue algae outbreak prediction result is obtained, including the following steps: The feature attention layer weight is calculated by combining the remote sensing-weather feature and the hydrology-time sequence correlation feature. According to the feature attention layer weight, the outbreak probability and time prediction are performed, and a blue algae outbreak prediction result is obtained.

10. A cyanobacterial bloom prediction system based on low-altitude multispectral remote sensing, characterized by, The following modules are included: A data acquisition module acquires low-altitude multi-spectral remote sensing monitoring data. A remote sensing feature parameter acquisition module uses the low-altitude multi-spectral remote sensing monitoring data to construct a blue algae core parameter inversion model, and obtains remote sensing feature parameters through the blue algae core parameter inversion model. A meteorological comprehensive influence index acquisition module performs spatio-temporal matching on meteorological data and the low-altitude multi-spectral remote sensing monitoring data, and obtains a meteorological comprehensive influence index according to the matching result. A hydrology dynamic influence index acquisition module realizes spatial interpolation of hydrology data through a hydrology dynamic model, and constructs a hydrology dynamic influence index according to the interpolation result. A blue algae growth feature parameter acquisition module introduces a dynamic cycle length, and obtains a blue algae growth feature parameter by combining the remote sensing feature parameter and the dynamic cycle length. A blue algae outbreak result prediction module fuses the remote sensing feature parameter, the meteorological comprehensive influence index, the hydrology dynamic influence index and the blue algae growth feature parameter, and obtains a blue algae outbreak prediction result.

11. An electronic device, comprising: It includes: One or more processors; A memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, Executable instructions are stored thereon, which are executed by a processor to make the processor implement the method of any one of claims 1 to 9.

Images