A benthic habitat assessment method and related devices based on cross-scale predictions of deep-sea indicator organisms

Abstract

The application discloses a kind of based on deep-sea indicator organism cross-scale prediction benthic habitat evaluation method and related equipment, method includes: obtaining the first data of training area and the second data of prediction area;Based on the first data, construct Gaussian process classification model, and then utilize the second data to predict output and obtain the existence probability prediction value of target indicator organism in prediction area;Based on the first data, construct Gaussian process regression model, and then utilize the second data to predict output and obtain the abundance data of target indicator organism in prediction area;Abundance data includes predicted abundance and abundance prediction variance;Existence probability prediction value and abundance data of prediction area are encoded and fused, and the comprehensive distribution map of prediction area is obtained;Based on the comprehensive distribution map, the function area of prediction area is divided using preset threshold condition.The present application provides direct scientific and technological support for formulating accurate deep-sea protection strategy and avoiding development risk, and can be widely applied in the field of data processing technology.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and related equipment for benthic habitat assessment based on cross-scale prediction of deep-sea indicator organisms. Background Technology

[0002] Deep-sea ecosystems play a crucial role in global carbon cycling and biodiversity maintenance, and the characterization of their benthic habitats is essential for understanding deep-sea ecological functions, assessing the impacts of environmental change, and developing conservation strategies. However, due to high survey costs (such as ROV / AUV exploration, seabed photography, and deep-sea trawls) and extreme environmental conditions (high pressure, low temperature, and darkness), deep-sea benthic habitat research faces core challenges such as high data sparsity, strong spatial heterogeneity, and mismatched observation scales. Summary of the Invention

[0003] The main objective of this invention is to propose a method, apparatus, electronic device, storage medium, and program product for benthic habitat assessment based on cross-scale prediction of deep-sea indicator organisms, aiming to solve at least one problem of the prior art.

[0004] To achieve the above objectives, one aspect of this invention proposes a method for benthic habitat assessment based on cross-scale prediction of deep-sea indicator organisms, the method comprising:

[0005] Obtain the first data for the training region and the second data for the prediction region;

[0006] The first data includes the first location information, first topographic factor, first environmental factor, first solid mineral coverage, and tag information and quantity information of target indicator organisms in the training area. The type of tag information includes presence or absence. The second data includes the second location information, second topographic factor, second environmental factor, and second solid mineral coverage of the prediction area.

[0007] A Gaussian process classification model is constructed based on the first data, and then the second data is used to predict the output to obtain the predicted probability value of the presence of the target indicator organism in the prediction area.

[0008] A Gaussian process regression model is constructed based on the first data, and then the abundance data of the target indicator organism in the prediction area is obtained by using the second data to predict the output; the abundance data includes the predicted abundance and the abundance prediction variance.

[0009] The predicted probability of existence in the predicted area is encoded and fused with the abundance data to obtain a comprehensive distribution map of the predicted area.

[0010] Based on the comprehensive distribution map, the predicted area is divided into functional regions using preset threshold conditions.

[0011] In some embodiments, obtaining first data for the training region includes the following steps:

[0012] Acquire seabed camera data, first multibeam sonar data, and first environmental observation data of the training area;

[0013] Based on the seabed camera data, the first location information, the first solid mineral coverage, and the tag information and quantity information of the target indicator organisms were extracted;

[0014] The first terrain factor is extracted based on the first multibeam sonar data, and the first environmental factor is extracted based on the first environmental observation data.

[0015] In some embodiments, obtaining second data for the prediction region includes the following steps:

[0016] Acquire second location information, second multibeam sonar data, and second environmental observation data for the predicted area;

[0017] In this case, the resolution of the second location information obtained from the prediction region is lower than the resolution of the first location information applied to the training region.

[0018] The second terrain factor is extracted based on the second multibeam sonar data, and the second environmental factor is extracted based on the second environmental observation data;

[0019] Based on the second topographic factor and the second environmental factor, the Kriging method is used to predict the second solid mineral coverage corresponding to the extracted second location information.

[0020] In some embodiments, a Gaussian process classification model is constructed based on the first data, and then the second data is used to predict the output to obtain the predicted probability value of the presence of the target indicator organism in the prediction area, including the following steps:

[0021] The label information, the first terrain factor, the first environmental factor, and the first solid mineral coverage are used as the first training data for the corresponding first location information;

[0022] The second topographic factor, the second environmental factor, and the second solid mineral coverage are used as the first predicted data for the corresponding second location information;

[0023] Based on the first training data, the spatial variability of habitat suitability in the training area is quantified as the target variance.

[0024] A Gaussian process classification model is obtained by combining the target variance with Euclidean distance and a preset length scale, and with a preset kernel function.

[0025] Among them, the Euclidean distance represents the measurement data between each location point in the training area. The Euclidean distance is calculated based on the latitude and longitude of each location point included in the first location information.

[0026] The first prediction data is input into the Gaussian process classification model, and the output is the predicted probability value of the presence of the target indicator organism in the prediction area.

[0027] In some embodiments, a Gaussian process regression model is constructed based on the first data, and then the abundance data of the target indicator organism in the prediction region is obtained by using the second data to predict the output, including the following steps:

[0028] The quantity information, the first topographic factor, the first environmental factor, and the first solid mineral coverage are used as the second training data corresponding to the first location information;

[0029] The second topographic factor, the second environmental factor, and the second solid mineral coverage are used as the second predicted data corresponding to the second location information;

[0030] Based on the second training data, a Gaussian process regression model is obtained by combining the squared exponential kernel and the white noise kernel.

[0031] The second prediction data is input into the Gaussian process regression model, and the predicted abundance and abundance prediction variance of the target indicator organisms in the prediction area are output.

[0032] In some embodiments, the predicted existence probability value and abundance data of the predicted region are encoded and fused to obtain a comprehensive distribution map of the predicted region, including the following steps:

[0033] The predicted abundance is standardized to obtain the predicted abundance distribution of target indicator organisms in the predicted region;

[0034] By performing an inverse transformation on the abundance prediction variance, the confidence level of the predicted abundance distribution of target indicator organisms in each part of the prediction area is obtained.

[0035] The predicted probability of existence, predicted abundance distribution, and confidence level of the predicted area are encoded into the topographic map of the predicted area to obtain a comprehensive distribution map.

[0036] The integrated distribution map simultaneously encodes three-dimensional information: the predicted probability of existence, the predicted abundance distribution, and the confidence level.

[0037] In some embodiments, based on a comprehensive distribution map, the predicted region is divided into functional regions using preset threshold conditions, including the following steps:

[0038] Areas with a predicted probability value greater than the first threshold, a predicted abundance distribution greater than the second threshold, and a confidence level greater than the third threshold are classified as high ecological value areas.

[0039] Areas where the predicted probability value is greater than the first threshold, the predicted abundance distribution is greater than the second threshold, and the confidence level is less than the fourth threshold are designated as restricted development areas.

[0040] Areas with predicted probability values ​​less than the first threshold, predicted abundance distribution less than the second threshold, and confidence levels less than the third threshold but greater than the fourth threshold are designated as development zones.

[0041] Areas with a confidence level less than the fourth threshold and that cannot be classified into any of the following types of areas: high ecological value area, restricted development area, and area that can be considered for development are classified as knowledge gap areas.

[0042] To achieve the above objectives, another aspect of the present invention proposes a benthic habitat assessment device based on cross-scale prediction of deep-sea indicator organisms, the device comprising:

[0043] The data acquisition module is used to acquire the first data of the training region and the second data of the prediction region;

[0044] The first data includes the first location information, first topographic factor, first environmental factor, first solid mineral coverage, and tag information and quantity information of target indicator organisms in the training area. The type of tag information includes presence or absence. The second data includes the second location information, second topographic factor, second environmental factor, and second solid mineral coverage of the prediction area.

[0045] The Gaussian process classification module is used to construct a Gaussian process classification model based on the first data, and then use the second data to predict the output to obtain the predicted probability value of the presence of the target indicator organism in the prediction area.

[0046] The Gaussian process regression module is used to construct a Gaussian process regression model based on the first data, and then use the second data to predict the abundance data of the target indicator organism in the prediction area; the abundance data includes the predicted abundance and the abundance prediction variance;

[0047] The coding fusion module is used to code and fuse the predicted existence probability values ​​and abundance data of the predicted area to obtain a comprehensive distribution map of the predicted area.

[0048] The functional zoning module is used to divide the predicted area into functional zones based on the comprehensive distribution map and using preset threshold conditions.

[0049] To achieve the above objectives, another aspect of the present invention provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned method.

[0050] To achieve the above objectives, another aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method.

[0051] To achieve the above objectives, another aspect of the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method.

[0052] The embodiments of the present invention include at least the following beneficial effects: The present invention provides a method, device, electronic device, storage medium, and program product for benthic habitat assessment based on cross-scale prediction of deep-sea indicator organisms. This scheme acquires first data of a training area and second data of a prediction area. The first data includes first location information, first topographic factors, first environmental factors, first solid mineral coverage, and tag information and quantity information of target indicator organisms in the training area. The tag information includes presence or absence. The second data includes second location information, second topographic factors, second environmental factors, and second solid mineral coverage of the prediction area. A Gaussian process classification model is constructed based on the first data, and then the predicted probability of the presence of target indicator organisms in the prediction area is obtained using the second data. A Gaussian process regression model is constructed based on the first data, and then the abundance data of target indicator organisms in the prediction area is obtained using the second data. The abundance data includes predicted abundance and abundance prediction variance. The predicted probability of presence and abundance data of the prediction area are encoded and fused to obtain a comprehensive distribution map of the prediction area. Based on the comprehensive distribution map, the prediction area is functionally divided using preset threshold conditions. This invention organically combines two types of questions: the presence of organisms (classification) and the abundance of organisms (regression). By constructing Gaussian process classification and regression models respectively and fusing their outputs (presence probability, predicted abundance, and variance), it can comprehensively and three-dimensionally assess habitats from two dimensions: "distribution range" and "population density," providing richer and more reliable decision-making basis for subsequent functional zoning. Furthermore, the Gaussian process model is a non-parametric Bayesian model, whose core advantage lies in not only providing predicted values ​​but also quantifying the uncertainty (variance) of the predictions. This perfectly matches the "sparse and unevenly distributed" characteristics of deep-sea data. The model exhibits high prediction confidence in areas with dense data points and larger variance in sparse data areas, thus honestly reflecting knowledge gaps, avoiding overfitting and blind prediction, and significantly improving the scientific rigor and reliability of the assessment results. Specifically, the method of this invention clearly distinguishes between the "training region" and the "prediction region," and its model design allows the extrapolation or interpolation of patterns learned from small-scale, high-precision data to large-scale, low-precision data regions. This directly addresses the challenge of "mismatched observation scales," significantly expanding the spatial scope of the assessment and reducing the cost of comprehensive surveys. The resulting "integrated distribution map" and "functional zoning" transform complex model predictions into intuitive and actionable management tools. This enables decision-makers to clearly identify high ecological value areas, risk areas, and knowledge gaps, providing direct scientific and technological support for developing precise deep-sea protection strategies and mitigating development risks. Attached Figure Description

[0053] Figure 1This is a schematic diagram of an implementation environment for a benthic habitat assessment method based on cross-scale prediction of deep-sea indicator organisms provided in an embodiment of the present invention;

[0054] Figure 2 This is a flowchart illustrating a benthic habitat assessment method based on cross-scale prediction of deep-sea indicator organisms provided in an embodiment of the present invention.

[0055] Figure 3 This is a schematic diagram of the unfolding process for obtaining the first data of the training region provided in an embodiment of the present invention;

[0056] Figure 4 This is a schematic diagram of the unfolding process for obtaining the second data of the prediction region provided in an embodiment of the present invention;

[0057] Figure 5 This is a schematic diagram illustrating the unfolded process of step S200 provided in an embodiment of the present invention;

[0058] Figure 6 This is a schematic diagram illustrating the unfolded process of step S300 provided in an embodiment of the present invention;

[0059] Figure 7 This is a schematic diagram illustrating the unfolded process of step S400 provided in an embodiment of the present invention;

[0060] Figure 8 This is a schematic diagram illustrating the unfolded process of step S500 provided in an embodiment of the present invention;

[0061] Figure 9 This is a schematic diagram of the benthic habitat assessment device based on cross-scale prediction of deep-sea indicator organisms provided in an embodiment of the present invention;

[0062] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of this invention; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this invention as detailed in the appended claims.

[0064] It is understood that the terms “first,” “second,” etc., used in this invention may be used herein to describe various concepts, but unless specifically stated otherwise, these concepts are not limited by these terms. These terms are used only to distinguish one concept from another. For example, first information may also be referred to as second information without departing from the scope of embodiments of the invention, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to determination” as used herein may be interpreted as “when…” or “when…” or “in response to determination.”

[0065] The terms “at least one,” “multiple,” “each,” “any,” etc., used in this invention, “at least one” includes one, two, or more than two; “multiple” includes two or more than two; “each” refers to each of the corresponding multiple; and “any” refers to any one of the multiple.

[0066] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0067] Among related technologies, deep-sea habitat research faces core challenges such as high data sparsity, strong spatial heterogeneity, and mismatched observation scales, due to high survey costs (such as ROV / AUV detection, seabed photography, and deep-sea trawls) and extreme environmental conditions (high pressure, low temperature, and darkness).

[0068] In view of this, this invention provides a method and related equipment for benthic habitat assessment based on cross-scale prediction of deep-sea indicator organisms. This method acquires first data for a training area and second data for a prediction area. The first data includes first location information, first topographic factors, first environmental factors, first solid mineral coverage, and tag information and quantity information of target indicator organisms in the training area. The tag information includes presence or absence. The second data includes second location information, second topographic factors, second environmental factors, and second solid mineral coverage in the prediction area. A Gaussian process classification model is constructed based on the first data, and then the predicted probability of the presence of target indicator organisms in the prediction area is obtained using the second data. A Gaussian process regression model is constructed based on the first data, and then the abundance data of target indicator organisms in the prediction area is obtained using the second data. The abundance data includes predicted abundance and abundance prediction variance. The predicted probability of presence and abundance data in the prediction area are encoded and fused to obtain a comprehensive distribution map of the prediction area. Based on the comprehensive distribution map, the prediction area is functionally divided using preset threshold conditions. This invention organically combines two types of questions: the presence of organisms (classification) and the abundance of organisms (regression). By constructing Gaussian process classification and regression models respectively and fusing their outputs (presence probability, predicted abundance, and variance), it can comprehensively and three-dimensionally assess habitats from two dimensions: "distribution range" and "population density," providing richer and more reliable decision-making basis for subsequent functional zoning. Furthermore, the Gaussian process model is a non-parametric Bayesian model, whose core advantage lies in not only providing predicted values ​​but also quantifying the uncertainty (variance) of the predictions. This perfectly matches the "sparse and unevenly distributed" characteristics of deep-sea data. The model exhibits high prediction confidence in areas with dense data points and larger variance in sparse data areas, thus honestly reflecting knowledge gaps, avoiding overfitting and blind prediction, and significantly improving the scientific rigor and reliability of the assessment results. Specifically, the method of this invention clearly distinguishes between the "training region" and the "prediction region," and its model design allows the extrapolation or interpolation of patterns learned from small-scale, high-precision data to large-scale, low-precision data regions. This directly addresses the challenge of "mismatched observation scales," significantly expanding the spatial scope of the assessment and reducing the cost of comprehensive surveys. The resulting "integrated distribution map" and "functional zoning" transform complex model predictions into intuitive and actionable management tools. This enables decision-makers to clearly identify high ecological value areas, risk areas, and knowledge gaps, providing direct scientific and technological support for developing precise deep-sea protection strategies and mitigating development risks.

[0069] It is understood that the benthic habitat assessment method based on cross-scale prediction of deep-sea indicator organisms provided by this invention can be applied to any computer device with data processing and computing capabilities, and this computer device can be various types of terminals or servers. When the computer device in the embodiments is a server, the server is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, the terminal can be a smartphone, tablet, laptop, or desktop computer, but it is not limited to these.

[0070] like Figure 1 The diagram shown is a schematic representation of an implementation environment provided by an embodiment of the present invention. (Refer to...) Figure 1 The implementation environment includes at least one terminal 102 and a server 101. The terminal 102 and the server 101 can be connected via a network, either wirelessly or via a wired connection, to complete data transmission and exchange.

[0071] Server 101 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0072] Additionally, server 101 can also be a node server in a blockchain network. Blockchain is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms.

[0073] Terminal 102 can be a smartphone, tablet computer, laptop computer, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. Terminal 102 and server 101 can be directly or indirectly connected via wired or wireless communication, and this embodiment of the invention does not impose any limitations.

[0074] For example, based on Figure 1The implementation environment shown in this embodiment of the invention provides a benthic habitat assessment method based on cross-scale prediction of deep-sea indicator organisms. The following description uses the application of this benthic habitat assessment method based on cross-scale prediction of deep-sea indicator organisms in server 101 as an example. It can be understood that this benthic habitat assessment method based on cross-scale prediction of deep-sea indicator organisms can also be applied to terminal 102.

[0075] Reference Figure 2 , Figure 2 This is an optional flowchart of the benthic habitat assessment method based on cross-scale prediction of deep-sea indicator organisms provided in the embodiments of the present invention. The execution subject of the benthic habitat assessment method based on cross-scale prediction of deep-sea indicator organisms can be any of the aforementioned computer devices (including servers or terminals). Figure 2 The method may include, but is not limited to, steps S100 to S500.

[0076] Step S100: Obtain the first data of the training region and the second data of the prediction region;

[0077] The first data includes the first location information, first topographic factor, first environmental factor, first solid mineral coverage, and tag information and quantity information of target indicator organisms in the training area. The type of tag information includes presence or absence. The second data includes the second location information, second topographic factor, second environmental factor, and second solid mineral coverage of the prediction area.

[0078] It should be noted that in some embodiments, such as Figure 3 As shown, acquiring the first data of the training area may include the following steps: S110, acquiring seabed camera data, first multibeam sonar data and first environmental observation data of the training area; S120, extracting first location information, first solid mineral coverage, and tag information and quantity information of target indicator organisms based on the seabed camera data; S130, extracting first topographic factors based on the first multibeam sonar data and extracting first environmental factors based on the first environmental observation data.

[0079] Specifically, this invention clarifies the methods for acquiring and extracting high-value training data, ensuring the reliability and richness of the basic data for model training. By integrating multi-source data such as seabed photography, multibeam sonar, and environmental observations, and accurately extracting location, topography, environment, seabed sediment (solid mineral coverage), and bio-tag information, high-quality, multi-dimensional training samples are provided for the subsequent construction of a high-performance Gaussian process model. This is a key prerequisite for the model to accurately learn habitat-biological relationships.

[0080] It should be noted that in some embodiments, such as Figure 4As shown, obtaining the second data for the prediction area may include the following steps: S140, obtaining the second location information, second multibeam sonar data, and second environmental observation data of the prediction area; wherein, the resolution of the second location information obtained for the prediction area is lower than the resolution of the first location information applied to the training area; S150, extracting the second terrain factor based on the second multibeam sonar data, and extracting the second environmental factor based on the second environmental observation data; S160, using the Kriging method to predict the second solid mineral coverage corresponding to the extracted second location information based on the second terrain factor and the second environmental factor.

[0081] Specifically, this invention specifies the processing method for prediction area data, particularly noting that the resolution of seabed camera data in the prediction area can be lower than that in the training area. Its beneficial effects are: significantly reducing the application cost and technical threshold for large-scale deep-sea prediction; allowing the use of lower-cost, more readily available wide-area low-resolution data as prediction input; and estimating key parameters (such as solid mineral coverage) using spatial interpolation methods such as Kriging, thus realizing the feasibility of effectively predicting vast unknown areas at a limited cost, significantly enhancing the engineering and practical value of the method of this invention.

[0082] For example, in some specific implementations, data acquisition and processing for the training region and the prediction region can be achieved as follows:

[0083] (1) Seabed camera data: This invention aims to manually identify individual sea cucumbers and solid mineral coverage in seabed cameras, and obtain the presence / absence tag A1 and quantity N1 information of indicator organisms (e.g., sea cucumbers) and their location information (e.g., latitude and longitude) L1; its resolution is high, for example 3m.

[0084] (2) Delineation of prediction area: Delineate the prediction range according to the research and application objectives, and extract the latitude and longitude L2 of the prediction area at a relatively low resolution, such as 3000m.

[0085] (3) Multibeam sonar data: Based on L1, terrain factors are extracted from multibeam sonar data, including depth, slope, terrain index and roughness, collectively referred to as S1; based on L2, terrain factors are extracted from multibeam sonar data or publicly available global terrain data, including depth, slope, terrain index and roughness, collectively referred to as S2.

[0086] (4) Environmental observation data (i.e. environmental factors): According to L1, bottom temperature, salinity, five major nutrients, water flow field and POC (Particulate Organic Carbon flux, which refers to the total amount of particulate organic carbon input from rivers into the ocean) flux are obtained from measured data, and are collectively referred to as E1; according to L2, bottom temperature, salinity, five major nutrients, water flow and POC flux are obtained from global public datasets, and are collectively referred to as E2.

[0087] (5) Solid mineral coverage: The solid mineral coverage C1 corresponding to the latitude and longitude L1 is extracted by manually identifying the seabed camera; the solid mineral coverage C2 in L2 is predicted and extracted by the Kriging method.

[0088] Step S200: Construct a Gaussian process classification model based on the first data, and then use the second data to predict the output to obtain the predicted probability value of the presence of the target indicator organism in the prediction area;

[0089] It should be noted that in some embodiments, such as Figure 5 As shown, step S200 may include the following steps: S210, using label information, a first terrain factor, a first environmental factor, and a first solid mineral coverage rate as the first training data corresponding to the first location information; S220, using a second terrain factor, a second environmental factor, and a second solid mineral coverage rate as the first prediction data corresponding to the second location information; S230, based on the first training data, quantifying the spatial variability of habitat suitability in the training area as the target variance; S240, based on the target variance, combining Euclidean distance and a preset length scale, and combining a preset kernel function to obtain a Gaussian process classification model; wherein, Euclidean distance represents the measurement data between each location point in the training area, and Euclidean distance is calculated based on the latitude and longitude of each location point included in the first location information; S250, inputting the first prediction data into the Gaussian process classification model, and outputting the predicted probability value of the presence of the target indicator organism in the predicted area.

[0090] For example, in some specific implementations, taking sea cucumbers as the target indicator organism, the probability prediction value can be achieved by using a Gaussian process classification model to predict the presence probability as follows:

[0091] Sea cucumber presence probability prediction / habitat suitability prediction: Predicting sea cucumber presence probability using Gaussian process classification. .

[0092] 1) Input training data: A1 indicating the presence / absence (0 / 1) of organisms; S1 and E1 corresponding to the latitude and longitude location L1; C1 solid mineral coverage, which is a standardized value of [A1,S1,E1,C1];

[0093] 2) Construct a Gaussian Process Classification (GPC) model:

[0094] The kernel function is set to the Matérn 3 / 2 kernel. The Matérn 3 / 2 kernel is better suited to local fluctuations than the RBF (infinitely differentiable) kernel, avoiding over-smoothing; it is also more robust than the Matern 5 / 2 (higher-order smoothing), making it suitable for small sample data. The RBF kernel (infinitely differentiable) tends to over-smooth the spatial-environment response, while the higher-order smoothing Matérn 5 / 2 kernel tends to be conservative for small samples. This invention preferably uses the Matérn 3 / 2 kernel, balancing local fluctuations with generalization robustness, and better reflecting the spatial correlation structure of deep-sea sea cucumber distribution.

[0095]

[0096] Parameter explanation: (Length scale): Controls the range of influence of environmental factors; (Variance): Quantifies the spatial variability of habitat suitability; Euclidean distance between two points. [1]

[0097] 3) Input prediction data: topographic factor S2 and environmental factor E2 corresponding to latitude and longitude location L2; solid mineral coverage C2, input as the standardized value of [S2,E2,C2];

[0098] 4) Output Results: In Python, the Gaussian process classification model (gaussian_process.GaussianProcessClassifier) ​​is called. Based on the training data, its model parameters, such as the kernel function, are adjusted. Training parameters can be set as needed, such as the maximum number of iterations during prediction (max_iter_predict) and parallel computation (n_jobs). Then, based on the prediction data, the predicted probability of organism presence is output, ultimately providing the predicted probability of sea cucumber presence at position L2. It can also be based on the obtained existence probability prediction value Generate a suitability distribution map.

[0099] Specifically, this invention optimizes the probabilistic prediction of binary events such as the presence or absence of organisms. By introducing a "target variance" that quantifies spatial variability and combining it with a preset kernel function, the model can fully consider the spatial autocorrelation and heterogeneity of environmental factors. Thus, when predicting the probability of existence, it can not only provide the possibility of "yes" or "no," but also reflect the spatial confidence level of the prediction conclusion, resulting in more robust results.

[0100] Step S300: Construct a Gaussian process regression model based on the first data, and then use the second data to predict the output to obtain the abundance data of the target indicator organism in the prediction area;

[0101] The abundance data includes predicted abundance and abundance prediction variance.

[0102] It should be noted that in some embodiments, such as Figure 6 As shown, step S300 may include the following steps: S310, using quantity information, first topographic factor, first environmental factor, and first solid mineral coverage as second training data corresponding to the first location information; S320, using second topographic factor, second environmental factor, and second solid mineral coverage as second prediction data corresponding to the second location information; S330, based on the second training data, combining a squared exponential kernel and a white noise kernel to obtain a Gaussian process regression model; S340, inputting the second prediction data into the Gaussian process regression model, and outputting the predicted abundance and abundance prediction variance of the target indicator organisms in the predicted area.

[0103] For example, in some specific implementations, abundance prediction based on a Gaussian process regression model can be achieved as follows:

[0104] 1) Input data: Marine parameter information N1; Topographic factor S1, environmental factor E1 and solid mineral coverage C1 corresponding to latitude and longitude location L1, input as standardized values ​​of [N1,S1,E1,C1];

[0105] 2) Construct a Gaussian Process Regression (GPR) model:

[0106] Construct a composite kernel function: RBF kernel + white noise kernel (WhiteKernel) to ensure the smoothness of the abundance distribution prediction and to take into account observation noise;

[0107]

[0108] RBF kernel (squared exponential kernel, which can use radial basis functions): ; Two data points (or feature vectors, samples) in the input space can be any two points in a multidimensional space. Representing the Euclidean distance between two vectors, it measures the "straight-line" distance between two points in multidimensional space. This distance is the core of kernel function calculation, quantifying the similarity between two points.

[0109] Function: To capture the smoothing spatial effects of environmental variables (such as the continuous influence of temperature gradients on abundance).

[0110] Hyperparameter: Length scale (Automatic learning during optimization);

[0111] White noise kernel: , The noise variance represents the power or magnitude of random noise present in the data. This noise is usually assumed to follow a mean of zero and a variance of 0. The Gaussian (normal) distribution, The larger the value, the stronger the random fluctuation or uncertainty in the data (e.g., large measurement error, or the phenomenon itself has a large degree of unpredictability). The smaller the value, the "cleaner" the data and the less noise. This represents the Kronecker delta function, if... = (That is, this is the same data point), then =1. If ≠ (That is, these are two different data points), then =0, which simulates uncorrelated, independent and identically distributed random noise (such as measurement error, random sensor disturbance, etc.).

[0112] Application example: Defining a kernel function in Python:

[0113] kernel=RBF(length_scale=1.0)+WhiteKernel(noise_level=0.1)

[0114] The structure consists of an RBF kernel (with an initial length scale of 1.0) and a white noise kernel (with an initial noise level of 0.1).

[0115] Purpose: To model observation noise (such as sampling error, small-scale variation, and avoid overfitting);

[0116] Hyperparameter: Noise variance ;

[0117] 3) Input prediction data: topographic factor S2 and environmental factor E2 corresponding to latitude and longitude location L2; solid mineral coverage C2, input as the standardized value of [S2,E2,C2];

[0118] 4) Output Results: In Python, the Gaussian process regression model `gaussian_process.GaussianProcessRegressor` is called. Based on the training data, its model parameters, such as the kernel function and the observation noise variance `alpha`, are adjusted. Kernel parameters can be set to optimize the number of restarts (`n_restarts_optimizer`) as needed. Then, based on the predicted data, the predicted biomass abundance and other information are output, ultimately providing the expected L2 sea cucumber abundance. Predicted sea cucumber abundance in the region The prediction variance of the uncertainty of the quantification prediction results .

[0119] Among them, the sea cucumber abundance distribution in the predicted area Output value of sea cucumber abundance for L2 With resolution coefficient k =(L2 resolution / L1 resolution) 2 The product of:

[0120]

[0121] For example, if the training region resolution is 100 meters and the prediction region resolution is 1000 meters, then the sea cucumber abundance in the prediction region... Output value of sea cucumber abundance for L2 *100.

[0122] Specifically, this invention optimizes the prediction of continuous values ​​for biodiversity (quantity) and uniquely provides a quantitative indicator of prediction uncertainty. The squared exponential kernel function effectively captures the complex nonlinear relationship between environmental factors and biodiversity, while the introduction of a white noise kernel effectively explains random noise or measurement errors in the data. The final output of "predicted abundance" and "abundance prediction variance" together constitute a complete estimate of the quantity. The variance value directly indicates the reliability of the prediction results, a crucial piece of information that traditional regression models cannot provide.

[0123] Step S400: Encode and fuse the predicted probability values ​​of the predicted area with the abundance data to obtain a comprehensive distribution map of the predicted area;

[0124] It should be noted that in some embodiments, such as Figure 7 As shown, step S400 may include the following steps: S410, standardizing the predicted abundance to obtain the predicted abundance distribution of target indicator organisms in the predicted area; S420, performing an inverse transformation on the abundance prediction variance to obtain the confidence level of the predicted abundance distribution of target indicator organisms at each location in the predicted area; S430, encoding the predicted existence probability value, predicted abundance distribution, and confidence level of the predicted area into the topographic map of the predicted area to obtain a comprehensive distribution map; wherein, the comprehensive distribution map simultaneously encodes the three-dimensional information of the predicted existence probability value, predicted abundance distribution, and confidence level.

[0125] For example, in some specific implementations, obtaining a comprehensive distribution map of the prediction region through encoding fusion can be achieved as follows:

[0126] A prediction result fusion mechanism is introduced, which weights the outputs of GPC and GPR, taking into account both "habitat suitability distribution" and "abundance level prediction", to obtain a comprehensive distribution map that combines "suitability-abundance-confidence". This map simultaneously encodes the following three-dimensional information:

[0127] 1) Habitat distribution suitability: the probability of sea cucumbers existing ;

[0128] 2) Sea cucumber predicted abundance index: Standardized sea cucumber predicted abundance distribution ;

[0129]

[0130] in, , This represents the expected abundance of all sea cucumbers. The minimum and maximum values ​​in.

[0131] 3) Confidence: Confidence level Variance can be predicted from abundance. The reverse conversion yields:

[0132]

[0133] in, Represents all predicted variances The maximum value in.

[0134] Specifically, this invention achieves multi-dimensional information integration and visualization of the evaluation results, greatly improving the interpretability and practicality of the findings. By simultaneously encoding the three layers of information—presence probability, standardized abundance, and confidence level—into a single topographic map, a comprehensive "three-dimensional" distribution map is generated. Decision-makers can clearly see where organisms might be present (presence probability), where organisms are abundant (abundance), and how confident they are in these judgments (confidence level). This fusion method preserves all the information from the model output to the greatest extent possible, avoiding information loss and laying a solid foundation for refined functional zoning.

[0135] Step S500: Based on the comprehensive distribution map, the predicted area is divided into functional areas using preset threshold conditions;

[0136] It should be noted that in some embodiments, such as Figure 8As shown, step S500 may include the following steps: S510, classifying areas with a predicted probability of existence greater than a first threshold, a predicted abundance distribution greater than a second threshold, and a confidence level greater than a third threshold as high ecological value areas; S520, classifying areas with a predicted probability of existence greater than the first threshold, a predicted abundance distribution greater than the second threshold, and a confidence level less than a fourth threshold as restricted development areas; S530, classifying areas with a predicted probability of existence less than the first threshold, a predicted abundance distribution less than the second threshold, and a confidence level less than the third threshold but greater than the fourth threshold as areas that can be considered for development; S540, classifying areas with a confidence level less than the fourth threshold and that cannot be classified into any of the high ecological value areas, restricted development areas, or areas that can be considered for development as knowledge gap areas.

[0137] For example, in some specific implementations, based on the "suitability-density-confidence" three-channel spectrum output by the Gaussian process model, the study area is divided into the following functional regions with clear management implications:

[0138] a) Areas with high ecological value and high suitability >0.7, high abundance >0.5, high confidence level The characteristic value is >0.8. This area is considered a core area for biodiversity and ecosystem function. Deep-sea mining and other activities should be mandated to avoid this area.

[0139] b) Restricted development areas (or areas requiring priority investigation) with high suitability >0.7, high abundance >0.5, low confidence region <0.5 characteristic. This area is considered to have the potential to become a high-ecological-value area, but due to insufficient current observational data, a definitive conclusion cannot be reached. Pre-emptive avoidance and proactive investigation should be conducted in this area. If the confidence level improves after investigation and it is confirmed as a high-value area ( If the ecological value is >0.8, it will be upgraded to a high ecological value area. If the investigation confirms that its ecological value is low (e.g., <0.7 or If the value is less than 0.5, it can be downgraded to a development zone that can be considered.

[0140] c) Development zones could be considered, which have low ecological suitability. <0.7 or <0.5, and with a moderate confidence level of 0.8> >0.5. The area has low known ecological value and biomass density, and the ecological damage from mining activities is expected to be within acceptable limits. Exploration and pilot mining can be permitted in this area under strict regulation. Even in such areas, the following must be implemented:

[0141] d) Knowledge gaps, low confidence level The value is <0.5 and cannot be classified into any of the above categories. A systematic baseline survey is needed for this area.

[0142] Specifically, this invention provides a scientific, quantitative, and repeatable method for regional management delineation, achieving a closed loop from scientific research prediction to management action. Specifically, by setting multi-parameter thresholds, subjective management decisions are transformed into objective, data-driven processes. Furthermore, it explicitly delineates "knowledge gap areas" (defined by low confidence levels) for the first time, which has significant practical implications. It can guide future scientific expeditions on which areas to prioritize for exploration to fill cognitive gaps, thereby optimizing resource allocation. Specifically, by distinguishing between "high ecological value areas" and "restricted development areas" (the latter being high-value but with insufficient understanding), it embodies the principle of prudent prevention, achieving a balance between protecting ecosystems and mitigating risks, making decisions regarding the protection and utilization of deep-sea resources more scientific, safe, and efficient.

[0143] It should also be noted that, in some optional implementations, the machine learning techniques used in these embodiments of the invention, besides Gaussian process classification and Gaussian process regression, can be extended using different mean functions, covariance functions (kernel functions), and combinations with different likelihood functions, such as deep Gaussian processes. Furthermore, depending on the biological distribution characteristics studied, the kernel of the Gaussian process in these embodiments of the invention can be adjusted or replaced.

[0144] Matérn 1 / 2 (ν=1 / 2): The corresponding process is not differentiable (too coarse).

[0145] Matérn 5 / 2 (ν=5 / 2): Higher-order smoothness, suitable for smoother functions.

[0146] Squared exponential kernel (RBF): can be viewed as a Matérn kernel as ν→∞, infinitely differentiable, and possibly oversmooth.

[0147] To explain in detail the principle of the technical solution of the present invention, the overall process of the present invention will be described below with reference to some specific embodiments. It is easy to understand that the following is an explanation of the technical principle of the present invention and should not be regarded as a limitation of the present invention.

[0148] First, it should be noted that the current research on habitat suitability and abundance level prediction has the following limitations: (1) Habitat suitability models based on binary classification (such as MaxEnt, GAMs, and random forests) rely only on species presence / absence data and cannot quantify biological abundance or biomass gradient, resulting in habitat suitability assessment being too binary (suitable / unsuitable); (2) Most commonly used abundance prediction models are abundance prediction models based on statistical regression (such as GLMs), assuming that species abundance and environmental factors have a globally smooth relationship, but deep-sea organisms (such as sea cucumbers) often show a highly patchy distribution; (3) Traditional habitat suitability and abundance level prediction research only outputs a single suitability / abundance prediction map; (4) The cross-scale prediction ability is weak. Deep-sea observation data is usually at the point scale (such as CTD water sampling and mooring sampling), while the model needs to predict the regional scale, and there is a scale effect deviation between the two.

[0149] Among these, seabed photography is the only efficient and non-destructive method for obtaining information on deep-sea benthic organisms and coverage. Sea cucumbers, belonging to the phylum Echinodermata, play a key ecological role in the ecosystem. In reflecting the characteristics of deep-sea benthic habitats, sea cucumbers have the following advantages: (1) They are large in size and have unique shapes, making them easy to identify in seabed photography; (2) They are widely distributed from the intertidal zone to the abyssal zone; (3) They have high population density; (4) They are sensitive to pollutants and environmental stress factors; (5) They have the ability to migrate short distances; (6) Compared with other marine organisms, they have a longer lifespan, which is conducive to long-term monitoring; (7) The density dependence and frequent patchy distribution of sea cucumbers increase their catchability and vulnerability, which may also exacerbate their sensitivity to deep-sea mining.

[0150] Currently, observational data such as seabed photography and publicly available data used in deep-sea habitat research are characterized by sparsity and high uncertainty. Therefore, there is a need to develop machine learning models for small sample sizes and uncertainty assessment. Bayesian methods, as a theoretically reliable mathematical tool, can consider noise in the system and uncertainty in the model during modeling, summarizing the information from the model and data into a posterior distribution, providing reasonable and easily interpretable estimates of uncertainty. Gaussian processes are a powerful non-parametric Bayesian modeling tool. They apply Bayesian inference to the function space, deriving a Gaussian posterior distribution from a Gaussian prior and observational data, thus simultaneously providing predictions and reliable estimates of their uncertainties. This has achieved good application results in solving global and regional deep-sea sediment and mineral resource prediction problems.

[0151] In light of this, this invention proposes for the first time to use the ecological characteristics of sea cucumbers as the core modeling basis, combining seabed camera data, multibeam echo information, and publicly available environmental data. It utilizes Gaussian process technology, which offers strong interpretability and uncertainty estimation capabilities, to establish a multi-scale habitat prediction system based on indicator species. Specifically, the Gaussian process regression model can define the smoothness and structure of the function through a kernel function (covariance function), and its flexibility can perfectly characterize patchy distributions. Furthermore, this invention's approach simultaneously considers intelligent methods for both "habitat suitability distribution" and "abundance level prediction," generating more complete, reliable, and ecologically significant prediction results. It produces a three-channel "suitability-density-confidence" map to support mining decisions in deep-sea basins lacking extensive observation.

[0152] Specifically, the benthic habitat assessment method based on cross-scale prediction of deep-sea indicator organisms provided by this invention can be implemented through the following process:

[0153] 1. Data Acquisition and Processing:

[0154] (1) Seabed camera data: This invention aims to manually identify individual sea cucumbers and solid mineral coverage in seabed cameras, and obtain the presence / absence tag A1 and quantity N1 information of indicator organisms (e.g., sea cucumbers) and their location information (e.g., latitude and longitude) L1; its resolution is high, for example 3m.

[0155] (2) Delineation of prediction area: Delineate the prediction range according to the research and application objectives, and extract the latitude and longitude L2 of the prediction area at a relatively low resolution, such as 3000m.

[0156] (3) Multibeam sonar data: Based on L1, terrain factors are extracted from multibeam sonar data, including depth, slope, terrain index and roughness, collectively referred to as S1; based on L2, terrain factors are extracted from multibeam sonar data or publicly available global terrain data, including depth, slope, terrain index and roughness, collectively referred to as S2.

[0157] (4) Environmental observation data (i.e. environmental factors): According to L1, bottom temperature, salinity, five major nutrients, water flow field and POC (Particulate Organic Carbon flux, which refers to the total amount of particulate organic carbon input from rivers into the ocean) flux are obtained from measured data, and are collectively referred to as E1; according to L2, bottom temperature, salinity, five major nutrients, water flow and POC flux are obtained from global public datasets, and are collectively referred to as E2.

[0158] (5) Solid mineral coverage: The solid mineral coverage C1 corresponding to the latitude and longitude L1 is extracted by manually identifying the seabed camera; the solid mineral coverage C2 in L2 is predicted and extracted by the Kriging method.

[0159] 2. Predictive model construction:

[0160] (1) Stage 1 (Prediction of sea cucumber presence probability / prediction of habitat suitability): Predicting the probability of sea cucumber presence using Gaussian process classification. :

[0161] 1) Input training data: A1 indicating the presence / absence (0 / 1) of organisms; S1 and E1 corresponding to the latitude and longitude location L1; C1 solid mineral coverage, which is a standardized value of [A1,S1,E1,C1];

[0162] 2) Construct a Gaussian Process Classification (GPC) model:

[0163] The kernel function is set to the Matérn 3 / 2 kernel. The Matérn 3 / 2 kernel is better suited to local fluctuations than the RBF (infinitely differentiable) kernel, avoiding over-smoothing; it is also more robust than the Matern 5 / 2 (higher-order smoothing), making it suitable for small sample data. The RBF kernel (infinitely differentiable) tends to over-smooth the spatial-environment response, while the higher-order smoothing Matérn 5 / 2 kernel tends to be conservative for small samples. This invention preferably uses the Matérn 3 / 2 kernel, balancing local fluctuations with generalization robustness, and better reflecting the spatial correlation structure of deep-sea sea cucumber distribution.

[0164]

[0165] Parameter explanation: (Length scale): Controls the range of influence of environmental factors; (Variance): Quantifies the spatial variability of habitat suitability; Euclidean distance between two points.

[0166] 3) Input prediction data: topographic factor S2 and environmental factor E2 corresponding to latitude and longitude location L2; solid mineral coverage C2, input as the standardized value of [S2,E2,C2];

[0167] 4) Output: Provide the predicted probability of the sea cucumber's presence at position L2. It can also be based on the obtained existence probability prediction value Generate a suitability distribution map.

[0168] (2) Stage 2 (Sea cucumber abundance prediction):

[0169] 1) Input data: Marine parameter information N1; Topographic factor S1, environmental factor E1 and solid mineral coverage C1 corresponding to latitude and longitude location L1, input as standardized values ​​of [N1,S1,E1,C1];

[0170] 2) Construct a Gaussian Process Regression (GPR) model:

[0171] Construct a composite kernel function: RBF kernel + white noise kernel (WhiteKernel) to ensure the smoothness of the abundance distribution prediction and to take into account observation noise;

[0172]

[0173] RBF kernel (squared exponential kernel, which can use radial basis functions): ;

[0174] Function: To capture the smoothing spatial effects of environmental variables (such as the continuous influence of temperature gradients on abundance).

[0175] Hyperparameter: Length scale (Automatic learning during optimization);

[0176] White noise kernel: ;

[0177] Purpose: To model observation noise (such as sampling error, small-scale variation, and avoid overfitting);

[0178] Hyperparameter: Noise variance ;

[0179] 3) Input prediction data: topographic factor S2 and environmental factor E2 corresponding to latitude and longitude location L2; solid mineral coverage C2, input as the standardized value of [S2,E2,C2];

[0180] 4) Output: Provide the expected abundance of sea cucumbers in L2. Predicted sea cucumber abundance in the region The prediction variance of the uncertainty of the quantification prediction results .

[0181] (3) Stage 3 (Prediction Result Output Layer):

[0182] A prediction result fusion mechanism is introduced, which weights the outputs of GPC and GPR, taking into account both "habitat suitability distribution" and "abundance level prediction", to obtain a comprehensive distribution map that combines "suitability-abundance-confidence". This map simultaneously encodes the following three-dimensional information:

[0183] 1) Habitat distribution suitability: the probability of sea cucumbers existing ;

[0184] 2) Sea cucumber predicted abundance index: Standardized sea cucumber predicted abundance distribution ;

[0185]

[0186] 3) Confidence: Confidence level Variance can be predicted from abundance. The reverse conversion yields:

[0187]

[0188] 3. To guide environmental management decisions related to deep-sea mining:

[0189] Based on the "suitability-density-confidence" three-channel spectrum output by the Gaussian process model, the study area is divided into the following functional regions with clear management implications:

[0190] a) Areas with high ecological value and high suitability >0.7, high abundance >0.5, high confidence level The characteristic value is >0.8. This area is considered a core area for biodiversity and ecosystem function. Deep-sea mining and other activities should be mandated to avoid this area.

[0191] b) Restricted development areas (or areas requiring priority investigation) with high suitability >0.7, high abundance >0.5, low confidence region <0.5 characteristic. This area is considered to have the potential to become a high-ecological-value area, but due to insufficient current observational data, a definitive conclusion cannot be reached. Pre-emptive avoidance and proactive investigation should be conducted in this area. If the confidence level improves after investigation and it is confirmed as a high-value area ( If the ecological value is >0.8, it will be upgraded to a high ecological value area. If the investigation confirms that its ecological value is low (e.g., <0.7 or If the value is less than 0.5, it can be downgraded to a development zone that can be considered.

[0192] c) Development zones could be considered, which have low ecological suitability. <0.7 or <0.5, and with a moderate confidence level of 0.8> >0.5. The area has low known ecological value and biomass density, and the ecological damage from mining activities is expected to be within acceptable limits. Exploration and pilot mining can be permitted in this area under strict regulation. Even in such areas, the following must be implemented:

[0193] d) Knowledge gaps, low confidence level The value is <0.5 and cannot be classified into any of the above categories. A systematic baseline survey is needed for this area.

[0194] In summary, this invention, for the first time, uses sea cucumbers (Echinodermata) as "indicator species" for deep-sea benthic habitats. It uses the ecological characteristics of sea cucumbers (such as easy identification, wide distribution, high population density, sensitivity to pollutants and environmental stressors, short-distance migratory ability, long lifespan, density dependence, and patchy distribution) as the core modeling basis. By utilizing their distribution patterns, it inversely determines the suitability and abundance of deep-sea habitats, establishing a multi-scale habitat indicator system. Furthermore, this invention, for the first time, combines Gaussian process classification and regression in deep-sea benthic habitat prediction, taking into account both suitability and abundance predictions. It also introduces uncertainty quantification technology, solving the problem that traditional habitat prediction models cannot accurately express prediction confidence. Specifically, this invention, through the fusion of seabed camera recognition, sonar topography, environmental observation, and publicly available big data, achieves a cross-scale prediction framework of local observation and regional prediction in the relatively undisturbed deep sea.

[0195] like Figure 9 As shown, this embodiment of the invention also provides a benthic habitat assessment device 900 based on cross-scale prediction of deep-sea indicator organisms, which can implement the above-described method. This device may include:

[0196] Data acquisition module 910 is used to acquire the first data of the training region and the second data of the prediction region;

[0197] The first data includes the first location information, first topographic factor, first environmental factor, first solid mineral coverage, and tag information and quantity information of target indicator organisms in the training area. The type of tag information includes presence or absence. The second data includes the second location information, second topographic factor, second environmental factor, and second solid mineral coverage of the prediction area.

[0198] Gaussian process classification module 920 is used to construct a Gaussian process classification model based on the first data, and then use the second data to predict the output to obtain the predicted probability value of the presence of the target indicator organism in the prediction area.

[0199] The Gaussian process regression module 930 is used to construct a Gaussian process regression model based on the first data, and then use the second data to predict the abundance data of the target indicator organism in the prediction area; the abundance data includes the predicted abundance and the abundance prediction variance;

[0200] The coding fusion module 940 is used to code and fuse the predicted existence probability value and abundance data of the predicted area to obtain a comprehensive distribution map of the predicted area.

[0201] The functional zoning module 950 is used to divide the predicted area into functional zones based on the comprehensive distribution map and using preset threshold conditions.

[0202] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0203] This invention also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0204] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0205] like Figure 10 As shown, Figure 10 The hardware structure of an electronic device 1000 according to another embodiment is illustrated. The electronic device 1000 includes:

[0206] The processor 1001 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (aSIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0207] The memory 1002 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RaM). The memory 1002 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1002 and is called and executed by the processor 1001.

[0208] Input / output interface 1003 is used to implement information input and output;

[0209] The communication interface 1004 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0210] Bus 1005 transmits information between various components of the device (e.g., processor 1001, memory 1002, input / output interface 1003, and communication interface 1004);

[0211] The processor 1001, memory 1002, input / output interface 1003 and communication interface 1004 are connected to each other within the device via bus 1005.

[0212] The electronic device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0213] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0214] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0215] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0216] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0217] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0218] The present invention provides a method, apparatus, electronic device, storage medium, and program product for benthic habitat assessment based on cross-scale prediction of deep-sea indicator organisms. This method acquires first data for a training area and second data for a prediction area. The first data includes first location information, first topographic factors, first environmental factors, first solid mineral coverage, and tag information and quantity information of target indicator organisms in the training area. The tag information includes presence or absence. The second data includes second location information, second topographic factors, second environmental factors, and second solid mineral coverage in the prediction area. A Gaussian process classification model is constructed based on the first data, and then the predicted probability of the target indicator organisms in the prediction area is obtained using the second data. A Gaussian process regression model is constructed based on the first data, and then the abundance data of the target indicator organisms in the prediction area is obtained using the second data. The abundance data includes predicted abundance and abundance prediction variance. The predicted probability of the prediction area and the abundance data are encoded and fused to obtain a comprehensive distribution map of the prediction area. Based on the comprehensive distribution map, the prediction area is functionally divided using preset threshold conditions. This invention organically combines two types of questions: the presence of organisms (classification) and the abundance of organisms (regression). By constructing Gaussian process classification and regression models respectively and fusing their outputs (presence probability, predicted abundance, and variance), it can comprehensively and three-dimensionally assess habitats from two dimensions: "distribution range" and "population density," providing richer and more reliable decision-making basis for subsequent functional zoning. Furthermore, the Gaussian process model is a non-parametric Bayesian model, whose core advantage lies in not only providing predicted values ​​but also quantifying the uncertainty (variance) of the predictions. This perfectly matches the "sparse and unevenly distributed" characteristics of deep-sea data. The model exhibits high prediction confidence in areas with dense data points and larger variance in sparse data areas, thus honestly reflecting knowledge gaps, avoiding overfitting and blind prediction, and significantly improving the scientific rigor and reliability of the assessment results. Specifically, the method of this invention clearly distinguishes between the "training region" and the "prediction region," and its model design allows the extrapolation or interpolation of patterns learned from small-scale, high-precision data to large-scale, low-precision data regions. This directly addresses the challenge of "mismatched observation scales," significantly expanding the spatial scope of the assessment and reducing the cost of comprehensive surveys. The resulting "integrated distribution map" and "functional zoning" transform complex model predictions into intuitive and actionable management tools. This enables decision-makers to clearly identify high ecological value areas, risk areas, and knowledge gaps, providing direct scientific and technological support for developing precise deep-sea protection strategies and mitigating development risks.

[0219] Specifically, the downstream product of the technical solution of this invention can be a program or software. Users can import datasets, classification parameters, and plotting parameters in the corresponding format through the user interface of the program or software. The program or software will automatically calculate and draw benthic habitat maps and provide interactive functions, enabling users to explore different habitat suitability and biodiversity distribution results and parameter settings. Based on this information, combined with the distribution characteristics of mineral resources, corresponding deep-sea mining habitat management strategies can be formulated, providing important scientific support for the protection and management of deep-sea ecosystems and promoting sustainable resource utilization and habitat protection.

[0220] The embodiments described in this invention are for the purpose of more clearly illustrating the technical solutions of the embodiments of this invention, and do not constitute a limitation on the technical solutions provided by the embodiments of this invention. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this invention are also applicable to similar technical problems.

[0221] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present invention, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0222] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0223] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0224] The preferred embodiments of the present invention have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and spirit of the present invention should be within the scope of the claims of the present invention.

Claims

1. A method for assessing benthic habitats based on cross-scale prediction of deep-sea indicator organisms, characterized in that, The method includes the following steps: Obtain the first data for the training region and the second data for the prediction region; The first data includes the first location information, first terrain factor, first environmental factor, first solid mineral coverage, and tag information and quantity information of the target indicator organism of the training area. The tag information includes the presence or absence of the target indicator organism. The second data includes the second location information, second terrain factor, second environmental factor, and second solid mineral coverage of the prediction area. Based on the first data, a Gaussian process classification model is constructed, and then the second data is used to predict the output to obtain the predicted probability value of the presence of the target indicator organism in the prediction region. A Gaussian process regression model is constructed based on the first data, and then the abundance data of the target indicator organism in the prediction region is obtained by using the second data to predict the output; the abundance data includes predicted abundance and abundance prediction variance; The predicted probability value of the predicted region is encoded and fused with the abundance data to obtain a comprehensive distribution map of the predicted region. Based on the comprehensive distribution map, the predicted area is divided into functional regions using preset threshold conditions.

2. The method according to claim 1, characterized in that, The process of obtaining the first data for the training region includes the following steps: Acquire seabed camera data, first multibeam sonar data, and first environmental observation data of the training area; Based on the seabed camera data, the first location information, the first solid mineral coverage, and the tag information and quantity information of the target indicator organism are extracted; The first terrain factor is extracted based on the first multibeam sonar data, and the first environmental factor is extracted based on the first environmental observation data.

3. The method according to claim 1, characterized in that, Obtaining the second data for the prediction region includes the following steps: Acquire the second location information, second multibeam sonar data, and second environmental observation data of the predicted area; Wherein, the resolution of the second location information obtained by the prediction region is lower than the resolution of the first location information applied by the training region; The second terrain factor is extracted based on the second multibeam sonar data, and the second environmental factor is extracted based on the second environmental observation data; Based on the second topographic factor and the second environmental factor, the Kriging method is used to predict the second solid mineral coverage corresponding to the extracted second location information.

4. The method according to claim 1, characterized in that, The process of constructing a Gaussian process classification model based on the first data, and then using the second data to predict the output to obtain the predicted probability value of the presence of the target indicator organism in the prediction region, includes the following steps: The label information, the first terrain factor, the first environmental factor, and the first solid mineral coverage are used as the first training data corresponding to the first location information; The second topographic factor, the second environmental factor, and the second solid mineral coverage are used as the first predicted data corresponding to the second location information; Based on the first training data, the spatial variability of habitat suitability in the training area is quantified as the target variance. Based on the target variance, combined with Euclidean distance and a preset length scale, and combined with a preset kernel function, the Gaussian process classification model is obtained. Wherein, the Euclidean distance represents the measurement data between each location point in the training area, and the Euclidean distance is calculated based on the latitude and longitude of each location point included in the first location information; The first prediction data is input into the Gaussian process classification model, and the predicted probability value of the presence of the target indicator organism in the prediction region is output.

5. The method according to claim 1, characterized in that, The process of constructing a Gaussian process regression model based on the first data, and then using the second data to predict the abundance data of the target indicator organism in the prediction region, includes the following steps: The quantity information, the first terrain factor, the first environmental factor, and the first solid mineral coverage are used as the second training data corresponding to the first location information; The second topographic factor, the second environmental factor, and the second solid mineral coverage are used as the second predicted data corresponding to the second location information; Based on the second training data, the Gaussian process regression model is obtained by combining the squared exponential kernel and the white noise kernel. The second prediction data is input into the Gaussian process regression model, and the predicted abundance and the predicted variance of the target indicator organism in the prediction region are output.

6. The method according to claim 1, characterized in that, The process of encoding and fusing the predicted existence probability value and the abundance data of the predicted region to obtain a comprehensive distribution map of the predicted region includes the following steps: The predicted abundance is standardized to obtain the predicted abundance distribution of the target indicator organism in the predicted region; The abundance prediction variance is inversely transformed to obtain the confidence level of the predicted abundance distribution of the target indicator organisms at each location in the prediction region; The predicted existence probability value, the predicted abundance distribution, and the confidence level of the predicted region are encoded into the topographic map of the predicted region to obtain the comprehensive distribution map. The integrated distribution map simultaneously encodes three-dimensional information of the existence probability prediction value, the predicted abundance distribution, and the confidence level.

7. The method according to claim 6, characterized in that, The step of dividing the predicted region into functional areas based on the comprehensive distribution map and using preset threshold conditions includes the following steps: Regions where the predicted probability of existence is greater than a first threshold, the predicted abundance distribution is greater than a second threshold, and the confidence level is greater than a third threshold are classified as high ecological value zones. Regions where the predicted probability of existence is greater than the first threshold, the predicted abundance distribution is greater than the second threshold, and the confidence level is less than the fourth threshold are classified as restricted development areas. The regions where the predicted probability of existence is less than the first threshold, the predicted abundance distribution is less than the second threshold, and the confidence level is less than the third threshold but greater than the fourth threshold are classified as development zones that can be considered. Areas with a confidence level less than the fourth threshold and that cannot be classified into any of the high ecological value areas, restricted development areas, or potential development areas are designated as knowledge gap areas.

8. A benthic habitat assessment device based on cross-scale prediction of deep-sea indicator organisms, characterized in that, The device includes: The data acquisition module is used to acquire the first data of the training region and the second data of the prediction region; The first data includes the first location information, first terrain factor, first environmental factor, first solid mineral coverage, and tag information and quantity information of the target indicator organism of the training area. The tag information includes the presence or absence of the target indicator organism. The second data includes the second location information, second terrain factor, second environmental factor, and second solid mineral coverage of the prediction area. The Gaussian process classification module is used to construct a Gaussian process classification model based on the first data, and then use the second data to predict the output to obtain the predicted probability value of the presence of the target indicator organism in the prediction area. The Gaussian process regression module is used to construct a Gaussian process regression model based on the first data, and then use the second data to predict the abundance data of the target indicator organism in the prediction region; the abundance data includes predicted abundance and abundance prediction variance; The encoding fusion module is used to encode and fuse the predicted existence probability value of the predicted region with the abundance data to obtain a comprehensive distribution map of the predicted region. The functional zoning module is used to divide the predicted area into functional regions based on the comprehensive distribution map and using preset threshold conditions.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images