An inland water body water quality multi-index remote sensing monitoring method based on a knowledge graph

Abstract

The application discloses a kind of inland water body water quality multi-index remote sensing monitoring methods based on knowledge graph, including the multi-source remote sensing image data of target water body area, water surface appearance and inherent optical characteristic data and historical water quality monitoring data;Using the remote sensing reflectivity-optical characteristic-water quality parameter joint data set obtained, the response relationship of water body optical characteristic and water quality component is analyzed, and the key information elements of each water quality parameter remote sensing inversion of water body are determined;Inland water body water quality multi-index remote sensing inversion knowledge graph is constructed, and the spatiotemporal continuity monitoring of water body multi-item water quality index is realized using remote sensing image.The present application is supplemented to remote sensing monitoring result by means of graph reasoning technology, and the completeness of monitoring data is significantly improved;Not only make up the deficiency of prior art in multi-index comprehensive monitoring and knowledge correlation expression, but also provide a new type of technical means and decision support for lake and river ecological environment management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of water quality monitoring, and particularly relates to a method for monitoring water quality of inland water bodies based on a knowledge graph. BACKGROUND

[0002] Under the background of global climate change and increasing human activities, the stability of inland water ecosystem is seriously threatened. Water quality monitoring of inland water bodies such as lakes and rivers is a key link in environmental protection and water resources management. Traditional water quality monitoring methods rely on manual sampling and laboratory analysis, which have high accuracy, but have problems such as long cycle, high cost, limited coverage, and are difficult to meet the needs of large-scale, high-frequency and dynamic change monitoring. In recent years, remote sensing technology has been widely used in water quality parameter monitoring due to its wide spatial coverage and strong timeliness. However, remote sensing monitoring can only obtain limited parameters of the surface layer of the water body, such as chlorophyll concentration, suspended matter concentration and water transparency, and relies on empirical modeling or statistical regression methods, which makes it difficult to fully explore the internal relationship and evolution law between different water quality indicators. In addition, most current remote sensing water quality monitoring methods focus on a single indicator, lack of multi-source heterogeneous data fusion and multi-indicator comprehensive analysis capability, which limits the systematicness and accuracy of the monitoring results.

[0003] In recent years, as an important research direction in the field of artificial intelligence, knowledge graph has been gradually introduced into environmental science due to its unique advantages in complex relationship modeling and knowledge organization. However, existing researches mainly focus on water environmental event tracing and pollution factor identification, and have not formed a systematic knowledge modeling framework for remote sensing water quality monitoring, and lack a mechanism for deep fusion of remote sensing data and multi-indicator water quality information. Therefore, it is urgent to build a new method for multi-indicator water quality monitoring combining remote sensing observation and knowledge graph. SUMMARY

[0004] The purpose of the present application is to provide a method for monitoring water quality of inland water bodies based on a knowledge graph.

[0005] The technical scheme of the present application is as follows:

[0006] (1) Collecting multi-source remote sensing image data, apparent and inherent optical property data, and historical water quality monitoring data of the target water body area;

[0007] (2) Using the obtained remote sensing reflectance-optical property-water quality parameter joint dataset to analyze the response relationship between water optical properties and water quality components, and to determine the key information elements for remote sensing inversion of each water quality parameter of the water body;

[0008] (3) Constructing the multi-index remote sensing inversion knowledge graph of the water quality of inland water bodies, and using remote sensing images to realize the spatiotemporal continuity monitoring of multiple water quality indexes of water bodies.

[0009] Further, the step (1) comprises:

[0010] (1.1) obtaining long-sequence remote sensing image data of the target water body and performing preprocessing to form a long-sequence remote sensing reflectance image and a water body inherent optical property image data set;

[0011] (1.2) obtaining water body apparent optical property and inherent optical property data of the target water body;

[0012] (1.3) obtaining water quality data of the target water body and performing data quality control;

[0013] (1.4) using the remote sensing reflectance image, the water body optical property data and the water quality data obtained in steps (1.1)-(1.3) to construct a remote sensing reflectance-optical property-water quality parameter joint data set.

[0014] Further, the preprocessing in step (1.1) comprises radiation correction, atmospheric correction, geometric correction, water body extraction and cloud detection.

[0015] Further, the step (2) comprises:

[0016] (2.1) using the inherent optical property and water quality parameter matching data in the remote sensing reflectance-optical property-water quality parameter joint data set to perform water body inherent optical property analysis;

[0017] (2.2) using the remote sensing reflectance, apparent optical property and water quality parameter pairing data in the remote sensing reflectance-optical property-water quality parameter joint data set to carry out quantitative analysis of the apparent optical property driving mechanism of the water body.

[0018] Further, the step (2.1) comprises:

[0019] (2.1.1) calculating the relative absorption contribution rates of total suspended particulate matter, pigment particulate matter, non-pigment particulate matter and colored dissolved organic matter in the water body, identifying the key components that dominate the change of the optical environment of the water body, and the calculation formula of the absorption contribution rate is as follows:

[0020] ,

[0021] wherein, represents the contribution rate of the i th component to the total absorption coefficient at the wavelength ; represents the absorption coefficient of the corresponding component at the wavelength , represents the total absorption coefficient at the corresponding wavelength;

[0022] (2.1.2) Based on the statistical relationship between the absorption and scattering coefficients of each optical component and the water quality index, the correlation analysis method is used to quantify the sensitivity intensity, which is represented as:

[0023] ,

[0024] wherein, r is the correlation coefficient, X represents the explanatory variable, Y represents the concentration of the water quality component of the lake, and are the average values, respectively.

[0025] Further, the step (3) comprises collecting optical signals, water quality signals, and inversion model information of inland water bodies, extracting key information of water quality remote sensing inversion of inland water bodies, quantitatively characterizing the relationship between water quality elements and key information, and applying remote sensing quantitative inversion model of inland water bodies.

[0026] Further, the collection of optical signals, water quality signals, and inversion model information of inland water bodies comprises:

[0027] (3.1) Remote sensing reflectance feature normalization, extracting remote sensing reflectance data matched with the target water body in the processed multi-source, multi-temporal remote sensing image, and marking according to the sensor type, time, region, and band;

[0028] (3.2) Water body optical parameter normalization, merging the water body apparent optical property and inherent optical property data of the corresponding sampling points to form different band absorption coefficients, scattering coefficients, and source substance compositions;

[0029] (3.3) Water quality monitoring index normalization, standardizing and annotating the spatial and temporal labels of each water quality index concentration in the measured water quality data;

[0030] (3.4) Integration and annotation of related data, integrating and matching the three types of information in steps (3.1)-(3.3) based on a unified time-space index, constructing a remote sensing reflectance-optical property-water quality index ternary

[0031] Joint database for knowledge graph modeling, and supplementing necessary semantic attribute labels;

[0032] (3.5) Proposing remote sensing inversion knowledge for the study of water body nutrient state index, including empirical methods, semi-empirical methods, and intelligent inversion algorithms.

[0033] Further, the extraction of key information of remote sensing inversion of inland water bodies comprises:

[0034] (3.6) Integrate the multi-source information knowledge collected in step (1) and step (2), screen the key information as entities in the knowledge graph, and select the remote sensing reflectivity, apparent optical properties, inherent optical properties and related water quality element concentrations that have strong correlation with each water quality index as the corresponding entities of each index according to the internal relationship between the water quality index and the water optical properties obtained in step (2.1), and clearly define their attribute definition, data type and semantic boundary;

[0035] (3.7) Extract the quantitative or qualitative relationship between entities from the physical analysis results;

[0036] (3.8) Rule pattern construction, extract the rule description in statistical law or empirical formula, and define it as a logical rule node for reasoning in the graph;

[0037] (3.9) Cross-modal link extraction, establish the semantic mapping relationship between different information types, and promote the knowledge graph to support intelligent retrieval and explanation across data modalities.

[0038] Further, the quantitative representation of the relationship between the water quality elements and the key information includes:

[0039] For each water quality index, the method described in the information collection step describes the mathematical relationship between the key information elements of the remote sensing inversion of each water quality parameter and the concentration level of the water quality index; all matching data are randomly divided into two groups, 70% for training the model, and the remaining 30% for verifying the application effect of the model, and ensuring that the training set and the validation set contain matching data of high, medium and low concentration levels, and the remote sensing inversion accuracy of each water quality index is evaluated; introduce time attribute and spatial identifier, support adaptive evolution and fusion of water quality knowledge in different times and regions, and construct knowledge graph query and visualization interface.

[0040] Further, the application of the inland water quality remote sensing quantitative inversion model includes:

[0041] (3.10) For the period covered by remote sensing images, combined with the response mechanism of remote sensing band-optical property-water quality index, sensitive band law, and dominant absorption component identification result knowledge triplets in the graph, construct an inversion rule path;

[0042] (3.11) For the period lacking remote sensing images, rely on the spatiotemporal law embedded in the graph, the evolution path between adjacent period data and known water quality parameters, and realize the intelligent supplement of the target period water quality index through the reasoning model based on the graph.

[0043] Beneficial effects: Compared with the prior art, the present application has the following significant advantages: the present application fuses remote sensing images and water quality observation data to construct a knowledge graph, expresses the correlation, evolution and influence mechanism between multi-index water quality factors through entity-relation modeling, and on this basis, the monitoring data completeness is significantly improved by means of graph reasoning technology to complete the remote sensing monitoring results; the method not only makes up for the deficiencies of the prior art in multi-index comprehensive monitoring and knowledge correlation expression, but also provides a new technical means and decision support for lake and river ecological environment management. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 The flowchart of the present application;

[0045] Figure 2 The remote sensing inversion knowledge diagram of each water quality index of inland water body. DETAILED DESCRIPTION

[0046] The technical solutions of the present application will be further described below in combination with the drawings.

[0047] As shown in Figure 1 The inland water body water quality multi-index remote sensing monitoring method based on knowledge graph according to the present application comprises the following steps:

[0048] (1) Collecting multi-source remote sensing image data, water body apparent and inherent optical property data and historical water quality monitoring data of the target water body area;

[0049] (1.1) Obtaining long sequence remote sensing image data of the target water body and performing preprocessing to form long sequence remote sensing reflectance image and water body inherent optical property image data set;

[0050] The remote sensing data includes multispectral or hyperspectral images such as Landsat, MODIS, MERIS, OLCI, GOCI, etc., which can be selected according to the spatial and temporal resolution requirements of water quality monitoring, monitoring index types, etc.

[0051] After obtaining the remote sensing image data, the original image needs to be preprocessed, mainly including radiation correction, atmospheric correction, geometric correction, water body extraction and cloud shadow removal, etc.

[0052] Radiation correction converts the digital signal received by the sensor into physical radiation brightness or radiance correction reflectance value, ensuring the comparability of data obtained at different times or by different sensors.

[0053] Atmospheric correction eliminates the influence of atmospheric scattering and absorption on the image, and can use a physical model-based method (such as 6S model) or an empirical linear method to form remote sensing reflectance image data that can be used for water quality monitoring.

[0054] Geometric correction, spatial registration of remote sensing images, to ensure that it is consistent with the geographical reference data or other data sources with the accuracy of spatial positioning.

[0055] Water extraction, which can be accurately separated from the water area in the remote sensing image by NDWI (water index), MNDWI and other methods, to provide a clear target area for subsequent water quality inversion.

[0056] Cloud detection, in order to ensure the accuracy of monitoring, cloud detection and removal is also needed, which often uses threshold-based or classification algorithm-based detection technology to ensure that the final image used for analysis is reliable and complete.

[0057] After the above steps of pretreatment, the remote sensing reflectance image data can be used to calculate the water inherent optical property index (including the absorption coefficient, scattering coefficient and backscattering coefficient of each water quality component of the water).

[0058] (1.2) Obtain the apparent optical property and inherent optical property data of the target water body;

[0059] Water apparent optical property data, i.e. off-water reflectance data, can be measured by ASSD FieldSpec Pro portable spectrometer of ASSD company using surface method to measure off-water reflectance data in 350-1050 nm wavelength range.

[0060] Water inherent optical property data, including absorption coefficient and backscattering coefficient of total suspended particles, pigment particles and non-pigment particles, and absorption coefficient of colored dissolved organic matter. Absorption coefficient can be measured by quantitative filter membrane technology, and the measurement range is 350-800 nm. Backscattering coefficient is measured by scattering instrument Hydroscat-6P produced by HOBI Labs company. Hydroscat-6P has 6 spectral channels in the visible to near infrared range, which are 442 nm, 488 nm, 532 nm, 590 nm, 767 nm and 852 nm.

[0061] In order to ensure the spatio-temporal matching of water optical data and water quality parameter data, water optical property parameters can be measured synchronously with water quality parameter data.

[0062] (1.3) Obtain water quality data of the target water body and perform data quality control;

[0063] Water quality monitoring data comes from ground monitoring stations or automatic monitoring equipment, including typical water quality parameters such as water chlorophyll, suspended solids, transparency, total nitrogen and total phosphorus suitable for remote sensing monitoring.

[0064] After data collection, data quality control is needed, mainly including missing value identification, outlier detection and consistency check. Missing value identification adopts time series integrity analysis, outlier detection combines statistical analysis (such as box plot analysis, Z-score method) with experience threshold to remove observation points obviously deviating from the normal range. Consistency check mainly through multi-site, multi-index cross comparison, to ensure the logical relationship and time trend of the data are reasonable. For individual data missing or abnormal but with recoverable, time interpolation, spatial interpolation method can be used for completion.

[0065] (1.4) Using the remote sensing reflectance image, water optical property data and water quality data obtained in steps (1.1)-(1.3), a remote sensing reflectance-optical property-water quality parameter joint data set is constructed.

[0066] After the collection and preprocessing of remote sensing images and water quality monitoring data are completed, spatio-temporal registration operation is needed to form remote sensing reflectance data consistent with ground water quality observation data in time and location.

[0067] Specifically, according to the observation time and spatial location of ground monitoring data, the remote sensing image pixel reflectance value matched therewith is extracted. In terms of time, by selecting the remote sensing observation time closest to the water quality sampling date (preferably the same day, if not, then select high-quality images within the previous 3 days), the correlation between reflectance characteristics and water quality status is ensured. In terms of space, according to the latitude and longitude coordinates of the monitoring site, the pixel value at the corresponding position is extracted from the remote sensing image using geographic information system (GIS) tools, if the site falls on the boundary of multiple pixels, then the matching accuracy can be enhanced by using the average value of surrounding pixels, weighted average or spatial interpolation, etc.

[0068] (2) Using the obtained remote sensing reflectance-optical property-water quality parameter joint data set, the response relationship between water optical properties and water quality components is analyzed, and the key information elements for remote sensing inversion of each water quality parameter of the water body are determined;

[0069] (2.1) Using the inherent optical property and water quality parameter matching data in the remote sensing reflectance-optical property-water quality parameter joint data set, the water inherent optical property analysis is carried out; including the following steps:

[0070] (2.1.1) Calculate the relative absorption contribution rate of total suspended particles, pigment particles, non-pigment particles and colored dissolved organic matter in the water body, identify the key components that dominate the change of water optical environment, and the calculation formula of absorption contribution rate is as follows:

[0071] ,

[0072] wherein, represents the relative absorption contribution rate of the i-th component in the water body at the wavelength of j. i represents the relative absorption contribution rate of the i-th component in the water body at the wavelength of j. The contribution rate of each component to the total absorption coefficient; The absorption coefficient of the corresponding component at the wavelength The absorption coefficient of the corresponding component at the wavelength The total absorption coefficient at the corresponding wavelength (i.e., the sum of the absorption coefficients of all components); commonly used sensitive wavebands include the blue waveband (about 440 nm), the green waveband (about 550 nm), and the red waveband (about 670 nm), etc.

[0073] By performing the above proportional calculation in multiple wavebands, the dominant degree of absorption of each component substance at different spectral positions is analyzed, and the most significant dominant optical component affecting the reflectivity of the water surface is identified in combination with the spectral response trend, thereby providing a basis for subsequent model construction and semantic relationship establishment of the atlas.

[0074] (2.1.2) Based on the statistical relationship between the absorption and scattering coefficients of each optical component (such as total suspended particulate matter, pigment particulate matter, non-pigment particulate matter, and colored dissolved organic matter, etc.) and the water quality index, the correlation analysis method is used to quantify the sensitivity intensity, and the correlation analysis method is represented as:

[0075] ,

[0076] wherein, r is the correlation coefficient, X represents the explanatory variable, Y represents the concentration of the lake water quality component, and are the average values thereof. r between -1 and 1, if r the value is positive, indicating a positive correlation, r the value is negative, indicating a negative correlation. Generally, it is considered that, r when |r|>0.6, the two variables have a strong correlation, 0.4<|r|<0.6, r when |r|<0.6, the correlation is moderate, and r when |r|<0.4, the correlation is weak.

[0077] (2.1.3) The relative absorption contribution rates of each component of the water body obtained above are sorted from high to low, and the correlation coefficients of the absorption, scattering coefficients of each component of the water body and the water quality index are sorted from strong to weak, and the inherent optical characteristic parameters with high relative absorption contribution rate and strong correlation coefficient are selected for subsequent water quality remote sensing inversion model construction.

[0078] (2.2) The quantitative analysis of the apparent optical property driving mechanism of the water body is carried out by using the remote sensing reflectivity, apparent optical property, and water quality parameter pairing data in the remote sensing reflectivity-optical property-water quality parameter joint data set.

[0079] including the following steps:

[0080] Firstly, the synchronous back-shrink method is used to screen the typical scenes of the collected water-leaving remote sensing reflectance or the paired samples of remote sensing reflectance and water quality indicators (such as chlorophyll a, total phosphorus, total nitrogen, etc.). This method is based on principal component analysis and statistical clustering techniques to retain representative band combinations and sampling scenes, and to eliminate redundant and unstable band information.

[0081] Then, the water-leaving reflectance or remote sensing reflectance data in the typical scenes screened is visualized as a function curve of the concentration of each water quality indicator (i.e. the band response graph is drawn with water quality concentration as the horizontal axis and different band reflectances as the vertical axis), which intuitively identifies the most sensitive band region to the concentration change of each water quality indicator.

[0082] Finally, based on the identification of sensitive bands, correlation analysis and regression analysis are performed between the reflectance of the corresponding band and each water quality indicator. Correlation analysis can use methods such as Pearson correlation coefficient, partial correlation analysis, mutual information analysis, etc. to quantify the response strength between sensitive bands and water quality indicators. Regression analysis can use methods such as linear, polynomial, support vector regression or semi-analytical models based on physical mechanisms, etc. to establish a mapping model from reflectance to water quality concentration, thereby analyzing the driving mechanism of remote sensing reflectance characteristics on water quality changes.

[0083] Through the above analysis process, the sensitive band region that dominates the apparent optical changes of the water body can be systematically identified, and the response relationship between the change of each band reflectance and the change of water quality indicators can be determined, thereby providing data support and physical interpretation for the subsequent construction of entity attributes and semantic relationship modeling of "reflectance-water quality indicators" in the water quality remote sensing inversion knowledge graph.

[0084] (3) Constructing an inland water quality multi-index remote sensing inversion knowledge graph to realize the spatiotemporal continuous monitoring of multiple water quality indicators in water bodies using remote sensing images. It mainly includes four steps: collection of information such as inland water optical signals, water quality signals, and inversion models, extraction of key information for inland water quality remote sensing inversion, quantitative representation of the relationship between water quality elements and key information, and application of inland water quality remote sensing quantitative inversion models.

[0085] The collection of information such as inland water optical signals, water quality signals, and inversion models includes:

[0086] According to the key information elements determined in step (2) for water quality parameter remote sensing inversion, the key information elements in remote sensing reflectance and water optical characteristics are sorted out as the data basis for water quality remote sensing monitoring.

[0087] (3.1) Remote sensing reflectance feature normalization, extract the remote sensing reflectance data matching the target water body from the processed multi-source and multi-temporal remote sensing images, and standardize the labeling according to the sensor type, time, region, and waveband;

[0088] (3.2) Water body optical parameter normalization, merge the apparent optical properties and inherent optical properties data of the corresponding sampling points to form the absorption coefficient, scattering coefficient and source material composition (such as pigment particulate matter, non-pigment particulate matter, CDOM, etc.) of different wavebands;

[0089] (3.3) Water quality monitoring index normalization, standardize and space-time label the concentration of each water quality index in the measured water quality data;

[0090] (3.4) Integration and labeling of related data, fuse and match the three types of information in steps (3.1)-(3.3) based on a unified time-space index, construct a remote sensing reflectance-optical property-water quality index triple

[0091] joint database for knowledge graph modeling, and supplement necessary semantic attribute labels (such as water area type, observation time, pollution level, etc.);

[0092] (3.5) Collect and sort the existing knowledge in the field of remote sensing inversion of each water quality index of inland water bodies and the field of remote sensing inversion for target water bodies, including inversion strategy, algorithm form, function form, feature waveband and combination method, etc. In addition, considering that the existing research presents inversion methods with poor universality, especially the quantitative inversion of indicators such as chlorophyll a, total nitrogen and total phosphorus is significantly affected by the differences in regional characteristics of water bodies, it is necessary to mine the response law of water body optical properties and water quality components, and propose remote sensing inversion knowledge for the nutritional status indicators of the study water body, such as the form of empirical / semi-empirical method and intelligent inversion algorithm, as shown in Figure 2 .

[0093] Key information extraction for remote sensing inversion of inland water quality includes:

[0094] (3.6) Integrate and process the multi-source information knowledge collected in steps (1) and (2), select the key information as entities in the knowledge graph, and according to the internal relationship between the water quality indicators and the water body optical properties obtained in step (2.1), select the remote sensing reflectance, apparent optical properties (such as water leaving reflectance), inherent optical properties (such as absorption and scattering coefficients of key component substances) and related water quality element concentrations (such as chlorophyll concentration when considering nitrogen and phosphorus concentration inversion) with strong correlation for each water quality indicator as the corresponding entity, and clearly define its attribute definition, data type and semantic boundary;

[0095] (3.7) Extract the quantitative or qualitative relationship between entities from the physical analysis results; for example, "Band B665 is highly sensitive to the change of chlorophyll a" "CDOM is the dominant absorption material affecting the reflectivity of blue-green band" "The scattering coefficient of total suspended particles is positively correlated with the total phosphorus concentration", which is converted into structured triples such as <B665, high correlation, chlorophyll a> or <CDOM, dominant absorption, blue-green band>;

[0096] (3.8) Rule pattern construction, extract the rule description in statistical law or empirical formula (such as "when the CDOM absorption coefficient is higher than the threshold A, the blue band reflectivity decreases significantly"), which is defined as a logical rule node used for reasoning in the graph;

[0097] (3.9) Cross-modal link extraction, establish the semantic mapping relationship between different information types, such as mapping "band combination features" and "water quality index concentration", and promote the knowledge graph to support intelligent retrieval and explanation across data modalities.

[0098] The quantitative characterization of the relationship between water quality elements and key information includes:

[0099] Quantitative expression of the relationship between entities. For each water quality index, the band and band ratio algorithm, baseline algorithm, feature index algorithm, and segmentation algorithm obtained through the information collection step, intelligent methods such as BP neural network (BP), support vector machine (SVR), and stepwise linear regression (SLR), and the proposed innovative method describe the mathematical relationship between the key information elements of remote sensing inversion of each water quality parameter and the concentration level of water quality index.

[0100] For analytical / semi-analytical methods, use collected remote sensing reflectivity, water optical information data, and water quality index matching data for parameter regionalization calibration; for intelligent methods, randomly divide all matching data into two groups, 70% for training model, and the remaining 30% for verifying the application effect of the model, and ensure that the training set and validation set contain high, medium, and low concentration levels of matching data. Determination coefficient (R2), relative root mean square error (RMSE), mean absolute error (MAE), and relative error (MRE) can be used as evaluation indicators to evaluate the remote sensing inversion accuracy of each water quality index.

[0101] At the same time, introduce time attribute and spatial identifier to support the adaptive evolution and fusion of water quality knowledge in different time and regions. And build knowledge graph query and visualization interface, support graph structure display, path query, band inversion path backtracking and other functions.

[0102] The application of inland water quality remote sensing quantitative inversion model includes:

[0103] (3.10) For the period of remote sensing image coverage, combined with the knowledge triplets of "remote sensing band-optical property-water quality index" response mechanism, sensitive band law, and dominant absorption component identification results in the atlas, the inversion rule path is constructed. For example, when a certain period of remote sensing image is input, the system can query the sensitive band, matched water type and water quality factor based on the atlas, and then call the pre-defined empirical formula or intelligent algorithm model in the atlas to realize the pixel-by-pixel inversion of the water quality index, and output the spatial distribution map.

[0104] (3.11) For the period of lack of remote sensing image, the present application relies on the spatio-temporal law embedded in the atlas, the evolution path between the adjacent period data and the known water quality parameters, and realizes the intelligent supplement of the water quality index of the target period through the reasoning model based on the atlas (such as path reasoning, graph embedding prediction, rule propagation, etc.). For example, the possible water quality state of the period without remote sensing coverage can be inferred based on the time series attributes of the nodes and the "water quality trajectory" of the historical similar scene, and the prediction error confidence is quantified. Especially in the background of frequent remote sensing lack of image in cloudy and rainy season and limited ground monitoring frequency, the present application can significantly improve the spatio-temporal continuous monitoring capability of inland lakes water quality.

Claims

1. A remote sensing method for multi-indicator water quality monitoring of inland water bodies based on knowledge graphs, characterized in that, Includes the following steps: (1) Collect multi-source remote sensing image data, water surface and inherent optical property data, and historical water quality monitoring data of the target water body area. Among them, the remote sensing data includes multispectral or hyperspectral images; water surface optical property data, i.e., water reflectance data; water inherent optical property data, including the absorption coefficient and backscattering coefficient of total suspended particulate matter, pigment particulate matter and non-pigment particulate matter, and the absorption coefficient of colored dissolved organic matter. (2) Using the obtained remote sensing reflectance-optical properties-water quality parameter joint dataset, analyze the relationship between water body optical properties and water quality component response, and determine the key information elements for remote sensing inversion of each water quality parameter of the water body; (3) Construct a knowledge graph of remote sensing inversion of multiple water quality indicators of inland water bodies, and use remote sensing images to realize the spatiotemporal continuous monitoring of multiple water quality indicators of water bodies; Step (2) includes: (2.1) Analyze the inherent optical properties of water bodies using the inherent optical properties and water quality parameter matching data in the remote sensing reflectance-optical properties-water quality parameter joint dataset; (2.2) Using the paired data of remote sensing reflectance, apparent optical properties and water quality parameters in the remote sensing reflectance-optical properties-water quality parameter joint dataset, a quantitative analysis of the driving mechanism of apparent optical properties of water bodies is carried out. Step (2.1) includes: (2.1.1) Calculate the relative absorption contribution rates of total suspended particulate matter, pigmented particulate matter, non-pigmented particulate matter, and colored dissolved organic matter in the water body, identify the key components that dominate the changes in the optical environment of the water body, and the calculation formula for the absorption contribution rate is as follows: ， in, Indicates the first i Class of components at wavelength The contribution rate of the component to the total absorption coefficient; Indicates the corresponding component at wavelength The absorption coefficient at that location, This represents the total absorption coefficient at the corresponding wavelength; (2.1.2) Based on the statistical relationship between the absorption and scattering coefficients of each optical component and water quality indicators, the sensitivity intensity is quantified using correlation analysis. The correlation analysis method is expressed as follows: ， in, r The correlation coefficient, X Represents explanatory variables. Y Represents the concentration of lake water components. and These are their average values; (2.1.3) Sort the relative absorption contribution rates of each water component obtained above from high to low, and sort the correlation coefficients of the absorption and scattering coefficients of each water component obtained above with the water quality index from strong to weak. Select the inherent optical characteristic parameters with high relative absorption contribution rates and strong correlation coefficients for subsequent water quality remote sensing inversion model construction.

2. The remote sensing monitoring method for multiple indicators of inland water quality based on knowledge graphs according to claim 1, characterized in that, Step (1) includes: (1.1) Acquire long-sequence remote sensing image data of the target water body and preprocess it to form a long-sequence remote sensing reflectance image and water body inherent optical property image dataset; (1.2) Obtain data on the surface optical properties and inherent optical properties of the target water body; (1.3) Obtain water quality data for the target water body and perform data quality control; (1.4) Using the remote sensing reflectance images, water optical property data and water quality data obtained in steps (1.1)-(1.3), construct a joint dataset of remote sensing reflectance-optical property-water quality parameters.

3. The remote sensing monitoring method for multiple indicators of inland water quality based on knowledge graphs according to claim 2, characterized in that, The preprocessing step (1.1) includes radiometric correction, atmospheric correction, geometric correction, water extraction, and cloud detection.

4. The remote sensing monitoring method for multiple indicators of inland water quality based on knowledge graphs according to claim 1, characterized in that, Step (3) includes the collection of optical signals, water quality signals, and inversion model information of inland water bodies, the extraction of key information for remote sensing inversion of inland water quality, the quantitative characterization of the relationship between water quality elements and key information, and the application of the quantitative inversion model for remote sensing of inland water quality.

5. The remote sensing monitoring method for multiple indicators of inland water quality based on knowledge graphs according to claim 4, characterized in that, The collection of inland water body optical signals, water quality signals, and inversion model information includes: (3.1) Remote sensing reflectance feature normalization: Extract remote sensing reflectance data that match the target water body from the processed multi-source and multi-temporal remote sensing images, and standardize and label them according to sensor type, time, region and band. (3.2) Water body optical parameters are consolidated and the surface optical properties and inherent optical properties of the water body at the corresponding sampling points are merged to form data including absorption coefficients, scattering coefficients and their source substances in different bands; (3.3) Water quality monitoring indicators are organized and the concentrations of various water quality indicators in the measured water quality data are standardized and labeled with spatiotemporal tags; (3.4) Data integration and annotation: The three types of information from steps (3.1) to (3.3) are fused and matched based on a unified time-space index to construct a ternary model of remote sensing reflectance, optical properties and water quality indicators for knowledge graph modeling. Combine databases and supplement with necessary semantic attribute tags; (3.5) Propose remote sensing inversion knowledge for trophic status indicators of water bodies, including empirical methods, semi-empirical methods and intelligent inversion algorithms.

6. The remote sensing monitoring method for multiple indicators of inland water quality based on knowledge graphs according to claim 4, characterized in that, The key information extraction for the remote sensing inversion of inland water quality includes: (3.6) Integrate and process the multi-source information knowledge collected in steps (1) and (2), and select the key information as entities in the knowledge graph. Based on the intrinsic relationship between water quality indicators and water body optical properties obtained in step (2.1), select remote sensing reflectance, water body surface optical properties, inherent optical properties and related water quality element concentrations that have a strong correlation with each water quality indicator as entities corresponding to each indicator, and clarify their attribute definitions, data types and semantic boundaries. (3.7) Extract quantitative or qualitative relationships between entities from the results of physical analysis; (3.8) Rule pattern construction: extract the regular descriptions from statistical laws or empirical formulas and define them as logical rule nodes for reasoning in the graph; (3.9) Cross-modal link extraction establishes semantic mapping relationships between different information types, promoting knowledge graphs to support intelligent retrieval and interpretation across data modalities.

7. The remote sensing monitoring method for multiple indicators of inland water quality based on knowledge graphs according to claim 4, characterized in that, The quantitative characterization of the relationship between water quality elements and key information includes: For each water quality indicator, the mathematical relationship between key information elements retrieved from remote sensing and the concentration levels of each water quality indicator is described through information collection steps. All matching data are randomly divided into two groups: 70% is used to train the model, and the remaining 30% is used to verify the model's application effect. Both the training and validation sets are guaranteed to contain matching data at high, medium, and low concentration levels. The accuracy of remote sensing retrieval for each water quality indicator is evaluated. Temporal attributes and spatial identifiers are introduced to support the adaptive evolution and fusion of water quality knowledge under different times and regions, and a knowledge graph query and visualization interface is constructed.

8. The remote sensing monitoring method for multiple indicators of inland water quality based on knowledge graphs according to claim 4, characterized in that, The application of the inland water quality remote sensing quantitative inversion model includes: (3.10) For periods covered by remote sensing images, the inversion rule path is constructed by combining the response mechanism of remote sensing bands-optical characteristics-water quality indicators, the pattern of sensitive bands, and the knowledge triplet of the identification results of dominant absorption components in the image. (3.11) For periods lacking remote sensing images, the water quality indicators for the target period are intelligently supplemented by relying on the spatiotemporal patterns embedded in the map, the evolution paths between adjacent time period data and known water quality parameters, and the map-based reasoning model.

Images