Non-feeding large-water-surface pearl mussel suitable-for-culture area and suitable-for-culture density intelligent prediction method and system

Abstract

The invention discloses a non-feeding type large water surface pearl mussel proper cultivation area and proper cultivation density intelligent prediction method and system, and the method comprises the steps: taking a target water body as a water source, and measuring the unit time water filtration rate of a mussel body under a dynamic condition in real time; synchronously acquiring the phytoplankton growth rate, the natural death rate, the water body retention time and the amount of phytoplankton at a water exchange part; on the basis of a remote sensing image covering a target water body, performing pixel-by-pixel operation on the image by using an optimal chlorophyll a concentration inversion model, outputting corresponding chlorophyll a concentration raster data, and further performing calculation to obtain pixel-level phytoplankton quantity; the pixel-level phytoplankton quantity and measured data are jointly input into the Herman ecological model, and the pearl mussel culture density is calculated for each pixel of the target water body. According to the method, dynamic measured parameters of phytoplankton, remote sensing inversion of the concentration of chlorophyll a and spatialization of an ecological model are combined, and spatialization, dynamic and intelligent prediction of the breeding density of the non-feeding type large-water-surface pearl mussels is achieved.

Description

Technical Field

[0001] This invention relates to the field of intelligent aquaculture technology, and more specifically, to an intelligent prediction method and system for suitable areas and stocking densities of large-scale pearl mussels that do not require feeding. Background Technology

[0002] Freshwater pearl farming is an important advantageous industry in my country, with pearl mussels (such as the triangular sail mussel) as the main farming target. Pearl mussels are typical filter-feeding mollusks with significant ecological filtration functions, effectively removing plankton, organic particles, and algae from the water, making them a typical environmentally friendly aquaculture species. However, freshwater pearl mussel farming in large open areas has long faced a prominent contradiction: traditional methods, which generally employ feeding-based methods like fertilization and algae cultivation, lead to frequent problems such as eutrophication, cyanobacterial blooms, and mussel diseases, hindering the realization of their ecological filtration advantages; while the shift to environmentally friendly "no-feeding" ecological farming avoids external pollution, the lack of precise and dynamic assessment methods for the carrying capacity of natural food (phytoplankton) in the water, coupled with long-term reliance on experience for stocking density, results in low farming efficiency, unstable pearl quality, and technological bottlenecks hindering the industry's transformation and upgrading.

[0003] Existing methods for assessing suitable stocking densities for shellfish in specific areas have systemic limitations when addressing the aforementioned needs, and these limitations are further amplified in the specific scenario of freshwater pearl mussels. First, the key physiological parameters relied upon for these assessments (such as filtration rate) are mostly derived from static laboratory measurements, failing to reflect the actual feeding behavior of pearl mussels under natural water flow, temperature, and food dynamics. Second, traditional ecological carrying capacity models often treat lakes as homogeneous wholes, outputting only the "total capacity" of the water body, while ignoring the high spatial heterogeneity of food resources in freshwater lakes and reservoirs caused by topography, wind fields, inflows and outflows, and human activities. This makes it difficult for the "holistic assessment, uniform stocking" model to avoid localized overstocking or understocking, and thus cannot support refined aquaculture layout and ecological risk management. Furthermore, existing technological approaches are fragmented: while suitability assessments based on remote sensing and GIS can achieve spatial visualization of environmental factors, they are essentially ratings of static habitat conditions and do not couple the core ecological process of "phytoplankton productivity - shellfish filter feeding consumption." Therefore, they can only answer "where might be suitable for aquaculture" and cannot provide a quantitative density of "how many can be aquacultured." On the other hand, while dynamic ecosystem models with complex mechanisms can simulate detailed processes, they have extremely high requirements for data input and computing resources, making them difficult to apply to rapid and operational assessments of many large water surface scenarios in my country.

[0004] Therefore, current stocking density prediction technologies have the following problems in the field of freshwater pearl mussel farming: technologies that pursue assessment accuracy rely on complex models and massive amounts of data, which are not practical enough; technologies that pursue implementation feasibility can only rely on rough experience or static assessment, which lack accuracy. Summary of the Invention

[0005] To address the core issues in existing non-feeding large-scale pearl mussel farming, such as reliance on experience for density control and the difficulty in balancing assessment accuracy and feasibility, this invention proposes an intelligent prediction method and system for suitable farming areas and densities for non-feeding large-scale pearl mussels. By constructing an integrated, coupled architecture deeply oriented towards freshwater pearl mussel ecological farming scenarios, this invention organically integrates three technical modules belonging to different fields: in-situ dynamic ecological parameter measurement, high-resolution remote sensing inversion, and spatial calculation of ecological models. This constructs a full-process intelligent prediction system from "dynamic parameter perception—remote sensing inversion—intelligent ecological model prediction—spatial visualization," ultimately achieving pixel-level, dynamic, and intelligent precise assessment of farming density, thus overcoming systemic obstacles that previous single-technology methods could not overcome.

[0006] To achieve the above objectives, in a first aspect, the present invention provides an intelligent prediction method for suitable culture areas and suitable stocking densities of large-scale pearl mussels that do not require feeding, comprising:

[0007] A continuous flow measurement device simulating natural water dynamic conditions was deployed in a representative area of ​​the target water body. Using the target water body as the water source, the filtration rate per unit time of the mussels under dynamic conditions was measured in real time to obtain the on-site filtration rate Cl. ff Simultaneously acquire phytoplankton growth rate μ, natural mortality rate m, and water residence time RT; calculate the phytoplankton abundance P at the water exchange point based on the chlorophyll a concentration. e Based on remote sensing imagery covering the target water body, a pre-constructed optimal chlorophyll a concentration inversion model is used to perform pixel-by-pixel calculations on the imagery, outputting chlorophyll a concentration raster data. The pixel-level phytoplankton content P is calculated based on the pixel-level chlorophyll a concentration. The pixel-level phytoplankton content P is then compared with the measured filtration rate Cl. ff Phytoplankton growth rate μ, natural mortality rate m, water retention time RT, and phytoplankton abundance P at water exchange points. e The Herman ecological model is used as a common input, and the culture density B of each pearl mussel in a pixel is calculated using the following formula. ff :

[0008]

[0009] This enables pixel-level quantitative assessment and spatial display of breeding density.

[0010] The aforementioned technical solution combines in-situ measurement of pearl mussel filtration rate, phytoplankton dynamic parameters, satellite remote sensing inversion of chlorophyll a concentration, and spatial calculation using ecological models to achieve spatial, dynamic, and intelligent prediction of the culture density of large-scale, non-feeding pearl mussels. It also enables spatiotemporal dynamic and intelligent assessment of the culture density of large-scale, non-feeding pearl mussels. Calculating culture density using in-situ measurement of pearl mussel filtration rate overcomes the limitations of traditional static laboratory tank measurements, avoiding the influence of factors such as hypoxia and particle sedimentation that may occur under still water conditions. This allows for a more realistic and dynamic reflection of the filter-feeding capacity of shellfish in natural aquatic environments, providing reliable key input parameters for predicting culture density. This technical solution innovatively applies culture density and model estimation techniques to freshwater areas, predicting the susceptibility and optimal culture density of the study area through a combination of remote sensing inversion and the ecological response of large-scale pearl mussels, and visualizing the prediction results.

[0011] Furthermore, in the continuous flow measurement device, live pearl mussels are placed in different water tanks. The water tanks have inlets and outlets connected to a water storage tank. The other end of the water storage tank is used to connect to the target water body to maintain flow rate, water temperature, and dissolved oxygen conditions similar to those of the actual target water body, so as to simulate the natural hydrodynamic environment.

[0012] Furthermore, the construction of the optimal chlorophyll a concentration inversion model includes: downloading Sentinel-2 L2A multispectral images covering the target water body and collecting measured chlorophyll a data with a time difference of no more than ±10 days from the images; extracting multispectral band reflectance corresponding to the measured chlorophyll a sampling points, constructing single-band and multi-band combination indices, and performing correlation analysis with the measured chlorophyll a concentration; selecting the feature parameters with the highest correlation to chlorophyll a concentration based on correlation ranking; selecting highly correlated single-band or band combination features as independent variables of the model, using the measured chlorophyll a concentration as the dependent variable, and establishing a chlorophyll a concentration inversion model using statistical regression or machine learning algorithms; calculating evaluation indicators through cross-validation to determine the optimal chlorophyll a concentration inversion model with the best predictive performance. Thus, this invention proposes a lightweight modeling path of "data-driven, optimal selection" to address the complex ecological characteristics of inland water bodies.

[0013] Furthermore, the downloaded multispectral images are first resampled, image stitched, band fusion and vector boundary clipping to generate a multiband fused image with uniform resolution that includes Band 2, Band 3, Band 4, Band 5, Band 6, Band 7, Band 8 and Band 8A. Then, the multispectral band reflectance corresponding to the measured sampling points of chlorophyll a is extracted.

[0014] Furthermore, the combined index includes band ratio, difference, and combinations of three and four bands.

[0015] Furthermore, the statistical regression algorithm is the optimal one selected based on the evaluation index from among univariate linear regression, quadratic polynomial regression, exponential regression, logarithmic regression, and power function regression algorithms.

[0016] Furthermore, the machine learning algorithm is the optimal one selected from random forest, support vector machine, and extreme gradient boosting algorithm based on evaluation metrics.

[0017] Furthermore, the evaluation indicators include the coefficient of determination, root mean square error, and mean absolute error.

[0018] Furthermore, the pearl mussel farming density B per pixel of the target water body was calculated. ff The cell-by-cell aquaculture density results are then mapped to a rasterized distribution map, and combined with time-series images to achieve dynamic density updates, threshold partitioning, and aquaculture potential analysis, outputting spatial prediction results of pearl mussel aquaculture density.

[0019] Secondly, this invention provides an intelligent prediction system for suitable areas and stocking densities of large-scale, non-feeding pearl mussels, used to implement the intelligent prediction method for suitable areas and stocking densities of large-scale, non-feeding pearl mussels as described above. The system includes: a continuous flow measurement device simulating hydrodynamic conditions in natural water bodies, deployed in a representative area of ​​the target water body; using the target water body as a water source, real-time measurement of the mussel's filtration rate per unit time under dynamic conditions to obtain the on-site filtration rate Cl. ff Simultaneously acquire phytoplankton growth rate μ, natural mortality rate m, and water residence time RT; calculate the phytoplankton abundance P at the water exchange point based on the chlorophyll a concentration. e ;Optimal chlorophyll a concentration inversion model: Perform pixel-by-pixel calculations on the Sentinel-2 image covering the target water body and output the corresponding chlorophyll a concentration raster data;

[0020] Herman ecological model: Pixel-level phytoplankton content P is calculated based on pixel-level chlorophyll a concentration; the pixel-level phytoplankton content P is then compared with the measured filtration rate Cl. ff Phytoplankton growth rate μ, natural mortality rate m, water retention time RT, and phytoplankton abundance P at water exchange points. e The Herman ecological model is used as a common input, employing the following formula:

[0021]

[0022] Calculate the pearl mussel farming density B for each pixel of the target water body. ff This enables pixel-level quantitative assessment of breeding density.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] This invention, by introducing in-situ filtration rate measurement technology based on on-site flowing water conditions for the first time in the freshwater fisheries field, overcomes the systematic errors caused by directly using marine shellfish parameters or static laboratory data. It fundamentally improves the reliability and scenario applicability of ecological model input data, and significantly enhances the authenticity and applicability of pearl mussel farming density results. Furthermore, by deeply coupling high-resolution feed distribution maps provided by remote sensing inversion with pixel-level calculations of the ecological model, it revolutionizes the granularity of farming capacity assessment from "the entire lake" to "each water area," realizing spatial, dynamic, and intelligent prediction of pearl mussel farming density. This allows managers to clearly see where feed is abundant and where feed is scarce and farming should be reduced or stopped, achieving a leap from "blindly uniform stocking" to "precise differentiated layout," and solving the industry problem of "macro-level assessment failing to achieve micro-level layout." Compared to traditional methods that rely on experience or static experiments, this method enables intelligent, dynamic assessment of the entire lake area throughout the year, significantly improving the accuracy and scientific validity of prediction results. It also visualizes the aquaculture density results, providing an intuitive spatial distribution map, thus offering decision support for scientific stocking and water management, and achieving sustainable ecological development.

[0025] This invention introduces in-situ dynamic measurement technology of filtration rate under continuous water flow conditions in large-scale pearl mussel farming without feeding, obtaining filter feeding parameters that truly reflect the natural ecological state and improving the authenticity and applicability of the model input data. At the same time, it combines multi-temporal remote sensing images to retrieve chlorophyll a concentration, realizing quantitative characterization of available food resources in the water body. The remote sensing retrieval results and measured ecological parameters are then input into the Herman ecological model to predict the optimal farming density pixel by pixel, achieving high-precision prediction of spatialized, dynamic, and intelligent farming density from ecological process parameters.

[0026] This invention transforms the continuous flow measurement device that simulates hydrodynamic conditions in natural waters from a traditional physiological research tool into a key data acquisition module for dynamic assessment of aquaculture capacity. This provides subsequent models with high-fidelity, localized core biological process parameters that cannot be replaced by traditional static parameters or empirical values ​​from literature, and is an important cornerstone for the accurate assessment achieved by this method.

[0027] This invention spatializes the point-scale or homogeneous aquatic ecological model proposed by Herman et al., originally used for conceptual studies. Utilizing the aforementioned pixel-level phytoplankton abundance P as the spatial driving force, the model's computational unit is decomposed from the "entire lake" to "each pixel." Simultaneously, the model's core physiological parameter Cl... ffDerived from in-situ, scenario-based measurements. This dual-driven model of "high spatial resolution environmental input + localized high-fidelity biological parameters" outputs a spatial distribution map of aquaculture density down to a scale of 10 m, which is unavailable through traditional methods. ff .

[0028] This invention achieves a balance between the accuracy of complex models and the efficiency of simple models: it avoids reliance on complex and data-hungry fluid dynamics-ecology coupling models, and innovatively strengthens a structurally simple Herman model through "locally measured key parameters + high-resolution remote sensing drive". This strategy unexpectedly reduces data requirements and computational costs while ensuring sufficient accuracy, making it possible to conduct rapid and periodic aquaculture density assessments on large water surfaces, filling a gap in practical tools in this field.

[0029] This innovative method constructs an integrated intelligent assessment system encompassing "dynamic parameter sensing—remote sensing inversion—intelligent ecological model prediction—spatial visualization," effectively overcoming multiple incompatibilities encountered when directly transplanting marine aquaculture assessment technologies to freshwater environments, including hydrological, optical, and biological challenges. Through localization and spatialization, it breaks through the limitations of traditional aquaculture density control, enabling high-precision prediction and dynamic updating of aquaculture density at the entire lake scale. This significantly reduces on-site monitoring costs and ecological disturbance risks, providing reliable technical support and decision-making basis for the scientific stocking, environmental protection, and efficient development of pearl mussel ecological aquaculture in large waters. The various links in this system mutually verify and support each other. For example, remote sensing inversion results can be verified using ground point data; the density distribution predicted by the model can guide sampling deployment; and long-term monitoring data can, in turn, optimize filtration rate parameters and the inversion model. This integrated, adaptive optimization technical loop has an overall value far exceeding the simple sum of individual technical units. Attached Figure Description

[0030] The invention, its features and advantages will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings.

[0031] Figure 1 This is a flowchart of the intelligent prediction method for suitable breeding areas and breeding densities of large-scale pearl mussels without feeding in this embodiment of the invention.

[0032] Figure 2 This is a schematic diagram of the continuous flow measuring device for the triangular sail mussel in an embodiment of the present invention;

[0033] Figure 3 This is a schematic diagram of the water distribution of Qianhe Lake and the distribution of sampling points corresponding to the chlorophyll a concentration sampling data of the lake in an embodiment of the present invention;

[0034] Figure 4This is a schematic diagram illustrating the correlation between the band reflectance corresponding to the band combination in the preprocessed remote sensing data and the chlorophyll a concentration sampling data in an embodiment of the present invention.

[0035] Figure 5 This is a scatter plot verifying the accuracy of the optimal fitting curve corresponding to the traditional statistical regression model for chlorophyll a concentration inversion in Qianhe Lake in this embodiment of the invention, i.e., a verification effect diagram of the logarithmic regression test set.

[0036] Figure 6 This is a scatter plot verifying the accuracy of the optimal fitting curve corresponding to the machine learning model for chlorophyll a concentration inversion in Qianhe Lake in this embodiment of the invention, i.e., the verification effect diagram of random forest training set vs. test set;

[0037] Figure 7 This is a schematic diagram of the spatial distribution of chlorophyll a concentration in Qianhe Lake in an embodiment of the present invention;

[0038] Figure 8 This is a schematic diagram of the intelligent prediction results of the culture density of Triangular Sail Mussels in Qianhe Lake in an embodiment of the present invention. Detailed Implementation

[0039] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but these are not intended to limit the scope of the invention.

[0040] In the following detailed description, numerous specific details are set forth to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that well-known algorithms and models are not shown in detail to avoid obscuring the spirit of the invention; and that the techniques not detailed in the following effect examples are readily available prior art.

[0041] Furthermore, the execution order of actions, steps, etc., involved in the systems and methods shown in the claims, specification, and drawings can be implemented in any order, unless a specific order is explicitly specified, and as long as the output of the preceding processing is not used in the subsequent processing.

[0042] See Figure 1 This invention provides a method and system for intelligent prediction of suitable areas and stocking densities for large-scale pearl mussels that do not require feeding, comprising:

[0043] First, a continuous flow measurement device simulating natural water dynamic conditions was deployed in a representative area of ​​the target water body. Using the target water body as the water source, environmental parameters such as water temperature, dissolved oxygen, and suspended particle concentration were kept close to natural conditions. The filtration rate per unit time of the mussels under dynamic conditions was measured in real time to obtain the on-site filtration rate Cl. ff .

[0044] More specifically, see Figure 2A continuous flow measurement device simulating natural hydrodynamics was set up at the monitoring point. This device used live *Triplophysa micrantha* mussels as experimental subjects, placing them in different flow tanks. Each flow tank had an inlet connecting to a water storage tank and an outlet for drainage. The other end of the water storage tank was connected to the target water body, maintaining a flow velocity close to that of the target water body. The filtration rate per unit time of the mussels under dynamic conditions was measured in real time, yielding the on-site filtration rate Cl. ff .

[0045] The filtration rate parameters obtained by the device accurately reflect the filtration capacity of shellfish under natural hydrodynamic conditions in the target water body. Spatiotemporal corrections are achieved through measurements of individuals at different times and sizes, serving as the core input for the ecological model. Specifically, in the temporal dimension, repeated measurements are conducted in representative seasons such as spring, summer, autumn, and winter to obtain the dynamic variation of filtration rate with key environmental factors, thereby establishing seasonal dynamic parameters. In the individual size dimension, shellfish are divided into small, medium, and large sizes according to shell length, and their average filtration rate is measured to obtain the corresponding classification functional parameters for the size structure of wild populations.

[0046] Simultaneously acquire phytoplankton growth rate μ, natural mortality rate m, and water residence time RT; calculate the phytoplankton abundance P at the water exchange point based on the chlorophyll a concentration. e ; where RT is determined through hydrological monitoring data, P e The phytoplankton growth rate μ and natural mortality rate m were calculated based on the chlorophyll a concentration at the water exchange point. These parameters were determined through on-site dilution and cultivation experiments. Experimental conditions, including temperature, light, and nutrient concentration, were controlled to simulate the natural aquatic environment, ensuring the ecological representativeness of the parameters.

[0047] Simultaneously, Sentinel-2 L2A multispectral imagery covering the target water body was downloaded, and chlorophyll a measured data was collected within ±10 days of the image time difference. The images were resampled, stitched, fused, and vector boundary clipped to generate a unified 10 m resolution multiband fused image containing Band 2, Band 3, Band 4, Band 5, Band 6, Band 7, Band 8, and Band 8A, ensuring accurate extraction of water body boundaries and stability of subsequent calculations. The selected bands are located in the visible to near-infrared "optical water" window, possessing effective penetration into water bodies, and their reflectivity is highly sensitive to changes in chlorophyll concentration (especially in the critical "red edge" region). Band 1 was excluded due to severe atmospheric scattering and low signal-to-noise ratio, while Bands 9 to 12 were either dedicated atmospheric sounding bands or, due to their short-wave infrared characteristics, were strongly absorbed by water bodies and could not carry water composition information. By actively discarding these invalid and interfering bands, the model input data was fundamentally purified, significantly improving the accuracy and stability of the subsequent chlorophyll a concentration inversion model.

[0048] Multispectral reflectance corresponding to measured sampling points was extracted, and single-band and multi-band combination indices (including band ratios, differences, three-band and four-band combinations, etc.) were constructed. Correlation analysis was performed with the measured chlorophyll a concentration, and the feature parameters with the highest correlation to chlorophyll a concentration were selected based on the correlation ranking. Statistical regression algorithms (such as univariate linear regression, quadratic polynomial regression, exponential regression, logarithmic regression, and power function regression) and machine learning algorithms (including random forest, support vector machine, and extreme gradient boosting) were used to establish a chlorophyll a concentration inversion model. The coefficient of determination R², root mean square error (RMSE), and mean absolute error (MAE) were calculated through cross-validation to determine the optimal chlorophyll a concentration inversion model with the best predictive performance.

[0049] Using the optimal chlorophyll a concentration inversion model, pixel-by-pixel calculations are performed on Sentinel-2 images covering the target water body to output the corresponding chlorophyll a concentration raster data. Chlorophyll a concentration directly characterizes phytoplankton biomass and can be used as a spatial index of phytoplankton biomass P (g / m³) to quantitatively describe the pixel-level spatial distribution dataset of available food resources in the water body.

[0050] Finally, the pixel-level phytoplankton content P and the measured filtration rate Cl were compared. ff Phytoplankton growth rate μ, natural mortality rate m, water retention time RT, and phytoplankton abundance P at water exchange points. e The Herman ecological model is used as a common input, employing the following formula:

[0051]

[0052] Calculate the pearl mussel farming density B for each pixel of the target water body. ff (g / m³), enabling pixel-level quantitative assessment of culture density. The optimal culture density (g / m³) of pearl mussels in each water area of ​​the target water body is calculated pixel by pixel, realizing dynamic assessment from ecological process parameters to culture density.

[0053] Furthermore, the aquaculture density results calculated pixel by pixel can be mapped to a rasterized distribution map, and combined with time-series images to achieve dynamic density updates, threshold partitioning, and potential analysis, outputting spatial prediction results of pearl mussel aquaculture density, providing intelligent decision support for ecological aquaculture planning and large-scale water management.

[0054] This invention also provides an intelligent prediction system for suitable areas and stocking densities of large-scale, non-feeding pearl mussels for implementing the aforementioned intelligent prediction method for suitable areas and stocking densities of large-scale, non-feeding pearl mussels. The system includes: a continuous flow measurement device simulating hydrodynamic conditions in natural water bodies, deployed in a representative area of ​​the target water body; using the target water body as a water source, real-time measurement of the mussel's filtration rate per unit time under dynamic conditions to obtain the on-site filtration rate Cl. ff Simultaneously acquire phytoplankton growth rate μ, natural mortality rate m, and water residence time RT; calculate the phytoplankton abundance P at the water exchange point based on the chlorophyll a concentration. e ;Optimal chlorophyll a concentration inversion model: Perform pixel-by-pixel calculations on the Sentinel-2 image covering the target water body and output the corresponding chlorophyll a concentration raster data;

[0055] Herman ecological model: Pixel-level phytoplankton content P is calculated based on pixel-level chlorophyll a concentration; the pixel-level phytoplankton content P is then compared with the measured filtration rate Cl. ff Phytoplankton growth rate μ, natural mortality rate m, water retention time RT, and phytoplankton abundance P at water exchange points. e The Herman ecological model is used as a common input, employing the following formula:

[0056]

[0057] Calculate the pearl mussel farming density B for each pixel of the target water body. ff This enables pixel-level quantitative assessment of breeding density.

[0058] Example

[0059] This embodiment uses a non-feeding, large-scale pearl mussel (Triangular sail mussel) aquaculture ecosystem as the implementation object. In-situ measurements and remote sensing inversion verification experiments were conducted in a typical lake (Qianhe Lake, Bengbu City, Anhui Province) to construct a dynamic prediction system for the optimal culture density of pearl mussels at the lake scale, achieving intelligent and ecological regulation of the aquaculture process. Qianhe Lake is a typical Class III shallow lake in the middle and lower reaches of the Yangtze River, and its trophic status is comprehensively affected by watershed agricultural runoff, rainfall replenishment, and surrounding aquaculture activities. The implementation period covers June 2024 to July 2025, reflecting the dynamic changes of the lake under different water temperature, light, and trophic conditions, providing a foundation for continuous monitoring and model verification throughout the year. Sentinel-2 remote sensing reflectance data was acquired synchronously with the chlorophyll a concentration data of Qianhe Lake. To ensure the temporal matching between the chlorophyll a concentration sampling data and the remote sensing imagery, the sampling time was set during the satellite transit period, thereby improving the correlation between the two sets of data.

[0060] The entire technical process includes three core steps: (1) Multi-temporal remote sensing inversion of chlorophyll a concentration in water and construction of spatial distribution of phytoplankton biomass; (2) In-situ measurement of pearl mussel filtration rate and related ecological parameters under continuous flow conditions; (3) Couple remote sensing results with ecological models to predict the pearl mussel farming density of Qianhe Lake pixel by pixel, so as to realize the spatial, dynamic and intelligent prediction of lake farming density.

[0061] See Figure 3 Ninety-nine fixed monitoring points were set up in the Qianhe Lake area. Representative sampling points were set up at inlets and outlets, aquaculture areas, and non-aquaculture areas, taking into account the hydrodynamic characteristics of the water body, to comprehensively cover various hydrodynamic and aquaculture conditions in the water area. The measured data are chlorophyll concentration data, including chlorophyll concentration sampling data and location data (including longitude and latitude) of each sampling point.

[0062] Field sampling includes two parts: water sample collection and in-situ ecological parameter measurement.

[0063] Continuous flow filtration rate measurement: A continuous flow measurement device simulating natural hydrodynamics is set up at the monitoring point (e.g., Figure 2 Live *Triplophysa philippinensis* mussels were used as experimental subjects. By maintaining a water flow velocity similar to that of a lake, the filtration rate per unit time of the mussels under dynamic conditions was measured in real time to obtain the on-site filtration rate Cl. ff This method overcomes the limitations of traditional laboratory static water tank measurements, avoiding the influence of factors such as hypoxia and particle sedimentation that may occur under still water conditions on the filtration rate measurement. It can more realistically and dynamically reflect the filter feeding capacity of shellfish in natural water environments, providing reliable key input parameters for predicting culture density.

[0064] Phytoplankton dynamic parameters were measured simultaneously: phytoplankton growth rate (μ), natural mortality rate (m), water retention time (RT), and phytoplankton abundance at the water exchange site (P). e Among them, RT was obtained by combining hydrological monitoring data, and P e The concentration of chlorophyll a at the water exchange point was used as the basis for measurement. All measurements were taken within ±10 days of the remote sensing image acquisition time to ensure the spatiotemporal synchronization of parameters and remote sensing data, thereby guaranteeing the accuracy and reliability of subsequent intelligent predictions of aquaculture density.

[0065] Download the Sentinel-2 L2A multispectral image of Qianhe Lake. The acquisition time should be within 10 days of the measured chlorophyll a concentration data. The specific data dates are shown in Table 1.

[0066] Table 1. Measured chlorophyll a concentration and the date of matching remote sensing image.

[0067]

[0068] Image preprocessing steps include band resampling to 10 m spatial resolution, multi-scene image stitching, and vector boundary clipping. Using the GDAL library and ArcGIS software, multi-band reflectance data corresponding to monitoring points were extracted from satellite remote sensing images to construct a paired dataset of chlorophyll a concentration and multispectral band reflectance, providing a training foundation for the subsequent establishment of a chlorophyll a concentration inversion model.

[0069] In this embodiment, correlation analysis is performed on the measured chlorophyll a concentration data and the single-band and combination indices of multispectral images to screen characteristic parameters highly correlated with chlorophyll a concentration. For example, in this embodiment, the characteristic parameters obtained by the above method are B5-B6, B5-B7, B5-B8A, B3 / B5, (B3-B5) / (B3+B5), B5 / B3, B5-B8, (B2-B5) / B3, (B2-B5) / (B2+B5), B5 / B2, and B2 / B5 (see [link to documentation]). Figure 4 In the figure, the vertical axis represents the absolute value |r| of the linear correlation coefficient between different bands / band combinations, used to characterize the strength of the linear correlation between them. The red dashed line corresponds to the critical value of the 0.05 level significance test; when |r| is greater than this dashed line, it indicates that the correlation between the corresponding bands / band combinations is statistically significant. Subsequently, the selected feature parameters are used as independent variables, and the measured value of chlorophyll a concentration is used as the dependent variable to construct a chlorophyll a concentration inversion model, providing a foundation for subsequent regression or machine learning modeling.

[0070] In the process of constructing the chlorophyll a concentration inversion model, the dataset was divided into a training set (70%) and a test set (30%). For traditional statistical regression methods (including univariate linear regression, quadratic polynomial regression, exponential regression, logarithmic regression, and power function regression), since they are only applicable to a single independent variable, the single band or band combination feature with the highest correlation (B5-B6 in this embodiment) was selected as the model independent variable, and the measured value of chlorophyll a concentration was used as the dependent variable to construct different forms of fitting models for the chlorophyll a concentration of Qianhe Lake.

[0071] Specifically, on the training set, the band combination index with the highest correlation is used as the independent variable, and the chlorophyll a concentration sampling data is used as the dependent variable to construct a scatter plot. Based on the scatter plot, univariate linear fitting, quadratic polynomial fitting, exponential fitting, logarithmic fitting, and power function fitting are performed respectively to obtain five different empirical formulas. On this basis, the coefficient of determination R is calculated on both the training and test sets. 2The root mean square error (RMSE) and mean absolute error (MAE) were used as evaluation metrics for model accuracy. In this embodiment, five inversion models and evaluation metrics were obtained on the training and test sets, and the performance of different models was compared. The results are shown in Table 2.

[0072] Table 2. Chlorophyll a Concentration Inversion Model and Coefficients of Determination Based on Empirical / Semi-Empirical Algorithms

[0073]

[0074] Based on the accuracy evaluation results on the test set, the best-performing fitted model was extracted from the univariate linear inversion model, the quadratic polynomial inversion model, the exponential inversion model, the logarithmic inversion model, and the power function inversion model. This model was then used as the statistical regression inversion model for chlorophyll a concentration in Qianhe Lake. (See also...) Figure 5 In this embodiment, the logarithmic regression model has the best prediction effect on the test set, and therefore it was selected as the optimal statistical regression inversion model for the chlorophyll a concentration of Qianhe Lake, and the corresponding optimal fitting curve diagram was obtained.

[0075] For machine learning methods (such as Random Forest (RF), Support Vector Machine (SVR), and Extreme Gradient Boosting (XGBoost)), since they support multiple independent variable inputs, multiple feature parameters related to chlorophyll a concentration (B5-B6, B5-B7, B5-B8A, B3 / B5, (B3-B5) / (B3+B5), B5 / B3, B5-B8, (B2-B5) / B3, (B2-B5) / (B2+B5), B5 / B2, B2 / B5) are used as independent variables, and the chlorophyll a concentration sampling data is used as the dependent variable to establish a training set. Based on this training set, models are trained using RF, SVR, and XGBoost methods respectively, resulting in multiple chlorophyll a concentration inversion models. The prediction performance of different models is compared, and the results are shown in Table 3.

[0076] Table 3. Chlorophyll a concentration inversion model and coefficient of determination based on machine learning regression method

[0077]

[0078] From the random forest, support vector machine, and extreme gradient boosting models, the performance of different models was comprehensively compared, and the inversion model with the best overall performance was selected as the inversion model for chlorophyll a concentration in Qianhe Lake. See also Figure 6 In this embodiment, the best-performing random forest model was selected as the chlorophyll a concentration inversion model for Qianhe Lake based on comprehensive evaluation indicators, and a schematic diagram of the corresponding optimal fitting curve was obtained.

[0079] Based on a comparison of traditional statistical regression methods and machine learning methods, the results show that machine learning algorithms outperform traditional regression methods in overall performance, with the RF model R... 2 With a value > 0.418 and lower RMSE and MAE than other models, it performed best on the validation samples and was identified as the optimal chlorophyll a concentration model. This model was used as the core input and applied to the subsequent inversion of chlorophyll a concentration in Qianhe Lake and the dynamic quantitative assessment of aquaculture density, providing high-precision and spatialized phytoplankton biomass data support for aquaculture density prediction.

[0080] After determining the optimal chlorophyll-a concentration retrieval model, it was loaded into a remote sensing image processing environment, and calculations were performed pixel-by-pixel on the fused image within the Qianhe Lake area. Specifically, the multispectral reflectance of each pixel in the fused image was used as an input parameter and substituted into the optimal retrieval model to obtain the predicted chlorophyll-a concentration value for the corresponding pixel. By sequentially calculating all pixels within the lake area, a chlorophyll-a concentration distribution dataset covering the entire lake region was obtained.

[0081] Furthermore, the distributed dataset undergoes a quality check to ensure that the model output values ​​are within a reasonable range for the aquatic environment, without negative values ​​or extreme outliers exceeding actual ecological significance. Based on this, the resulting data is standardized, including: unifying the raster value type of the output data (e.g., float 32), maintaining a spatial reference system consistent with the input imagery, and specifying the unit of chlorophyll a concentration (mg / m³). This step differs from image preprocessing before modeling; its purpose is to ensure that the inversion results, as independent data products, can be directly applied to subsequent phytoplankton biomass calculations and aquaculture density assessments, and facilitate sharing, overlaying, and analysis within a GIS environment.

[0082] Finally, the processed dataset is rasterized and output as a chlorophyll a concentration distribution map (see [link]). Figure 7 This distribution map visually reflects the spatial heterogeneity of chlorophyll a concentration in Qianhe Lake. More importantly, it serves as key input data for phytoplankton biomass calculation and pearl mussel farming density assessment, providing core support for subsequent estimation of farming density and generation of spatialized farming density distribution maps.

[0083] After obtaining the chlorophyll a concentration distribution map within the lake area, the chlorophyll a concentration (μg / L) was used as the phytoplankton biomass P (g / m³). 3The phytoplankton biomass is a key pigment in phytoplankton photosynthesis, and its concentration is significantly positively correlated with phytoplankton biomass, directly reflecting the current quantity and abundance of phytoplankton. Therefore, in this embodiment, the retrieved chlorophyll a concentration results are used to express the spatial distribution of phytoplankton biomass. This yields a phytoplankton biomass distribution dataset covering the entire lake, realizing the transformation from "optical indicators" to "ecological quantitative indicators."

[0084] The pixel-level phytoplankton content P and the measured filtration rate Cl ff Phytoplankton growth rate μ, natural mortality rate m, water retention time RT, and phytoplankton abundance P at water exchange points. e The Herman ecological model is used as a common input, employing the following formula:

[0085]

[0086] Calculate the pearl mussel farming density B for each pixel of the target water body. ff (g / m 3 ).

[0087] Subsequently, the model calculations are performed pixel by pixel to obtain B covering the entire lake area. ff Spatial distribution dataset. The output data is normalized, including standardizing the raster resolution, spatial reference system, and numerical units, to ensure consistency with the chlorophyll a distribution map data.

[0088] Finally, B ff The dataset output is a spatial distribution map of aquaculture density (see...). Figure 8 This system enables dynamic quantitative assessment and visualization of pearl mussel farming density across the entire lake. The distribution map intuitively reveals the spatial differences in farming density across different water areas within the lake, providing technical support for scientifically formulating farming layouts, predicting farming density, and managing farming. It also establishes, for the first time, a dynamic assessment system for the farming density of large-scale dormant pearl mussels. Breaking through the limitations of traditional experience, this system can be implemented and promoted in different shellfish farming scenarios, providing a reliable basis for the intelligent management and green development of large-scale shellfish ecological farming.

[0089] While remote sensing and in-situ monitoring technologies have mature applications in their respective fields, integrating them across domains and adapting them to the unique ecological constraints of freshwater pearl mussel farming (such as low salinity, differentiated algal communities, and complex hydrology) requires overcoming a series of technical obstacles, from parameter localization and data fusion to scale matching. The goal is to construct a technical solution that balances "assessment accuracy" and "implementation feasibility," thereby addressing the long-standing industry pain points of "inaccurate measurement, incomplete visualization, and imprecise calculation." This invention constructs an integrated technical system of "dynamic parameter perception—remote sensing inversion—intelligent ecological model prediction—spatial visualization," achieving high-precision prediction and dynamic updating of farming density at the entire scale of Qianhe Lake. This significantly reduces on-site monitoring costs and ecological disturbance risks, providing reliable technical support and decision-making basis for the scientific stocking, environmental protection, and efficient development of large-scale pearl mussel ecological farming.

[0090] The method can be further embedded with an automated algorithm framework based on Python or cloud computing platform to realize intelligent modular operation of chlorophyll a time series inversion and dynamic updating of culture density. It has online data update and self-learning optimization capabilities, thereby realizing continuous optimization of pearl mussel culture density and intelligent decision support in large waters.

[0091] The aforementioned intelligent prediction method for suitable breeding areas and densities of large-scale pearl mussels without feeding can be embodied in the form of a computer program product or software functional unit. If this method is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Therefore, the essence of this technical solution, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an electronic system (which may be a personal computer, server, or network system, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0092] Those skilled in the art will recognize that the units, i.e., algorithm steps, of the various examples described in connection with this embodiment can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0093] The preferred embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above. Systems and structures not described in detail should be understood as being implemented in a conventional manner in the art. Any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention using the methods and techniques disclosed above, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the present invention. This does not affect the essential content of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the present invention are still within the scope of protection of the present invention.

Claims

1. A method for intelligently predicting the suitable culture area and stocking density of large-scale pearl mussels that do not require feeding, characterized in that, include: A continuous flow measurement device simulating the hydrodynamic conditions of natural water bodies was deployed in a representative area of ​​the target water body. Using the target water body as the water source, the filtration rate per unit time of the mussels under dynamic conditions was measured in real time to obtain the on-site filtration rate Cl. ff Simultaneously acquire phytoplankton growth rate μ, natural mortality rate m, and water residence time RT; The phytoplankton content (P) at the water exchange point was calculated based on the chlorophyll a concentration at the water exchange point. e ; Based on remote sensing images covering the target water body, the pre-constructed optimal chlorophyll a concentration inversion model is used to perform pixel-by-pixel calculations on the images to output chlorophyll a concentration raster data; the pixel-level phytoplankton content P is calculated based on the pixel-level chlorophyll a concentration. The pixel-level phytoplankton content P and the measured filtration rate Cl ff Phytoplankton growth rate μ, natural mortality rate m, water retention time RT, and phytoplankton abundance P at water exchange points. e The Herman ecological model is used as a common input, and the culture density B of each pearl mussel in a pixel is calculated using the following formula. ff : ， This enables pixel-level quantitative assessment and spatial display of breeding density.

2. The intelligent prediction method for suitable culture areas and stocking densities of large-scale pearl mussels without feeding, as described in claim 1, is characterized in that... In the continuous flow measurement device, live pearl mussels are placed in different water tanks. The water tanks have inlets and outlets connected to a water storage tank. The other end of the water storage tank is used to connect to the target water body to maintain flow rate, water temperature, and dissolved oxygen conditions that are similar to those of the actual target water body, so as to simulate the natural hydrodynamic environment.

3. The intelligent prediction method for suitable culture areas and stocking densities of large-scale pearl mussels without feeding, as described in claim 1, is characterized in that... The steps for constructing the optimal chlorophyll a concentration inversion model include: downloading Sentinel-2 L2A multispectral images covering the target water body and collecting measured chlorophyll a data within ±10 days of the image time difference; extracting multispectral band reflectance corresponding to the measured chlorophyll a sampling points, constructing single-band and multi-band combination indices, and performing correlation analysis with the measured chlorophyll a concentration; selecting the feature parameters with the highest correlation to chlorophyll a concentration based on correlation ranking; selecting highly correlated single-band or band combination features as independent variables of the model, using the measured chlorophyll a concentration as the dependent variable, establishing the chlorophyll a concentration inversion model using statistical regression or machine learning algorithms, calculating evaluation indicators through cross-validation, and determining the optimal chlorophyll a concentration inversion model with the best predictive performance.

4. The intelligent prediction method for suitable culture areas and stocking densities of large-scale pearl mussels without feeding, as described in claim 3, is characterized in that... The downloaded multispectral images are first resampled, image stitched, band fusion and vector boundary clipping to generate a multiband fused image with uniform resolution that includes Band 2, Band 3, Band 4, Band 5, Band 6, Band 7, Band 8 and Band 8A. Then, the multispectral reflectance corresponding to the measured sampling points of chlorophyll a is extracted.

5. The intelligent prediction method for suitable culture areas and suitable stocking densities of large-scale pearl mussels without feeding, as described in claim 3, is characterized in that... The combined index includes band ratio, difference, and combinations of three and four bands.

6. The intelligent prediction method for suitable culture areas and suitable stocking densities of large-scale pearl mussels without feeding, as described in claim 3, is characterized in that... The statistical regression algorithm is the optimal one selected based on the evaluation index from among univariate linear regression, quadratic polynomial regression, exponential regression, logarithmic regression, and power function regression algorithms.

7. The intelligent prediction method for suitable culture areas and stocking densities of large-scale pearl mussels without feeding, as described in claim 3, is characterized in that... The machine learning algorithm mentioned is the optimal one selected from random forest, support vector machine and extreme gradient boosting algorithm based on evaluation metrics.

8. The intelligent prediction method for suitable culture areas and suitable stocking densities of large-scale pearl mussels without feeding, as described in claim 3, is characterized in that... The evaluation indicators include the coefficient of determination, root mean square error, and mean absolute error.

9. The intelligent prediction method for suitable culture areas and stocking densities of large-scale pearl mussels without feeding, as described in claim 1, is characterized in that... The pearl mussel farming density B per pixel of the target water body was calculated. ff The cell-by-cell aquaculture density results are then mapped to a rasterized distribution map, and combined with time-series images to achieve dynamic density updates, threshold partitioning, and aquaculture potential analysis, outputting spatial prediction results of pearl mussel aquaculture density.

10. A smart prediction system for suitable culture areas and stocking densities of large-scale pearl mussels that do not require feeding, characterized in that, The method for implementing the intelligent prediction method for suitable breeding areas and breeding densities of large-scale pearl mussels without feeding as described in any one of claims 1 to 9 includes: A continuous flow measurement device simulating hydrodynamic conditions in natural water bodies: deployed in a representative area of ​​the target water body; using the target water body as the water source, the filtration rate per unit time of the mussels under dynamic conditions is measured in real time to obtain the on-site filtration rate Cl. ff Simultaneously acquire phytoplankton growth rate μ, natural mortality rate m, and water retention time RT; calculate the phytoplankton abundance P at the water exchange point based on the chlorophyll a concentration. e ; Optimal chlorophyll a concentration inversion model: Perform pixel-by-pixel calculations on Sentinel-2 images covering the target water body and output the corresponding chlorophyll a concentration raster data. Herman ecological model: Pixel-level phytoplankton content P is calculated based on pixel-level chlorophyll a concentration; the pixel-level phytoplankton content P is then compared with the measured filtration rate Cl. ff Phytoplankton growth rate μ, natural mortality rate m, water retention time RT, and phytoplankton abundance P at water exchange points. e The Herman ecological model is used as a common input, employing the following formula: ， Calculate the pearl mussel farming density B for each pixel of the target water body. ff This enables pixel-level quantitative assessment of breeding density.

Images