Lake trophic state stratified detection method based on laser radar and multispectral remote sensing

Abstract

The application discloses a kind of based on laser radar and multi-spectral remote sensing's lake nutrition state layered detection method, belong to the field of water quality measurement and analysis by means of laser radar and multi-spectral remote sensing data.For the technical problems such as insufficient precision of traditional lake multi-spectral remote sensing water depth detection, uneven sample caused by artificial isobathic stratification, the application extracts high-precision water depth points using adaptive density clustering model coupled with local topographic slope, estimated water depth and along-track photon density, detects global continuous water depth of lake combined with multi-spectral remote sensing data, adaptively divides non-overlapping water depth subspace based on water depth statistical distribution, constructs and optimizes the optimal water quality parameter detection model for each subspace, finally detects global water quality parameters and completes lake comprehensive nutrition state evaluation.The application realizes collaborative detection of multi-source remote sensing data, significantly improves the precision of water depth measurement and water quality detection, can integrally detect five core water quality parameters, and is suitable for large-scale lake water environment dynamic monitoring.

Description

Technical Field

[0001] This invention relates to the field of water quality measurement and analysis using lidar and multispectral remote sensing data, specifically to a method for stratified detection of lake trophic status based on lidar and multispectral remote sensing. Background Technology

[0002] Eutrophication of lakes is one of the most prominent water environment problems in my country and even globally. Accurately obtaining the spatial distribution of lake depth and water quality parameters (such as chlorophyll a, total phosphorus, total nitrogen, transparency, and permanganate index) is crucial for assessing the trophic status of lakes. Traditional water quality monitoring relies on manual point sampling and laboratory analysis, which has inherent drawbacks such as high cost, small coverage, and poor timeliness, making it difficult to meet the needs of large-scale, high-frequency, and refined dynamic monitoring of lakes.

[0003] Currently, multispectral remote sensing-based lake water quality monitoring technology has been widely used, but this technology still faces many technical bottlenecks:

[0004] On the one hand, multispectral remote sensing data itself is difficult to directly obtain high-precision lake water depth information. On the other hand, the optical properties of water bodies are closely related to water depth, and there are significant differences in the light transmission paths and light attenuation patterns between shallow and deep water areas. Most existing methods establish a unified detection model for the entire lake area, without fully considering the impact of differences in the spatial distribution of water depth on the spectral response of water quality parameters. This results in inconsistent detection accuracy in different water depth areas, and the overall error is difficult to meet the requirements of refined monitoring.

[0005] On the other hand, most existing multispectral water quality detection methods do not fully consider the impact of water depth on the spectral response of water quality parameters. Even those methods that attempt to model water depth stratification generally use manual division at equal depth intervals. This division method lacks data support and cannot guarantee the sufficiency and balance of training samples for each water depth interval. It is difficult to accurately adapt to the spectral response patterns of water quality parameters under different water depth conditions, resulting in generally low accuracy in overall water quality detection. In addition, most existing methods can only detect single or a few water quality indicators such as chlorophyll a and total phosphorus, and cannot form an integrated detection and evaluation system for the five core water quality indicators that meet national standards.

[0006] LiDAR technology, as an active remote sensing technology, can achieve high-precision water depth measurement by emitting laser pulses and receiving echo signals, providing a new approach to solving the problem of insufficient accuracy in multispectral remote sensing water depth detection. However, existing lidar water depth extraction techniques still have significant shortcomings in practical applications. LiDAR water echo signals are easily interfered with by various factors such as water turbidity, lakebed topography, and surface fluctuations, resulting in low accuracy in distinguishing between surface photons and lakebed photons. This leads to significant deviations in water depth measurement results, making it difficult to use as high-precision water depth calibration data for multispectral remote sensing water depth detection training.

[0007] Therefore, how to improve the accuracy and physical consistency of lake depth measurement and effectively integrate it into the stratified detection framework of water quality parameters in order to achieve accurate evaluation of lake trophic status is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0008] To alleviate or partially alleviate the above-mentioned technical problems, the solution of the present invention is as follows:

[0009] On one hand, this invention discloses a method for stratified detection of lake trophic status based on lidar and multispectral remote sensing, comprising:

[0010] Acquire lidar data and use an adaptive density clustering model that couples local terrain slope, estimated water depth and photon density along the track to identify lake surface photon clusters and lake bottom photon clusters in the lidar data, and obtain water depth point data after refraction correction.

[0011] Based on the aforementioned water depth point data and multispectral remote sensing data, the continuous water depth distribution characteristics of the entire lake area are obtained.

[0012] Based on the continuous water depth distribution characteristics of the entire lake area, the lake is divided into non-overlapping water depth subspaces based on statistical distribution characteristics.

[0013] For each of the aforementioned water depth subspaces, a water quality parameter detection model is constructed and trained, specifically including: for each of the aforementioned water depth subspaces, based on the original band reflectance of the multispectral remote sensing data, constructing spectral ratio features, difference features, and normalized difference features; and using measured water quality data, selecting the feature combination with the highest accuracy from a full-spectrum feature set containing the original band reflectance, the spectral ratio features, the difference features, and the normalized difference features to train the water quality parameter detection model; and

[0014] Based on the measured water quality data, the water quality parameter detection model with the highest accuracy is selected as the optimal detection model for this water depth subspace; and

[0015] Based on the optimal detection model for each water depth subspace, the distribution of water quality parameters across the entire lake area is calculated, and indicators for evaluating the comprehensive trophic status of the lake are obtained.

[0016] In one preferred embodiment, the adaptive density clustering model dynamically calculates the adaptive neighborhood radius and the adaptive minimum core point number for each photon point to be clustered, based on the local terrain slope, estimated water depth, and along-track photon density within the local calculation window where the photon point is located, and performs clustering based on the adaptive neighborhood radius and the adaptive minimum core point number.

[0017] In one preferred embodiment, the formula for calculating the adaptive neighborhood radius is:

[0018] ,

[0019] Where, ε i For adaptive neighborhood radius, ε base The base radius, k1 and k2 are adjustment coefficients, and θ slope D represents the local slope angle within the local calculation window. est To estimate the water depth, D max This represents the maximum depth of the lake.

[0020] In one preferred embodiment, the formula for calculating the adaptive minimum number of core points is:

[0021] ,

[0022] Among them, MinPts i To achieve the minimum number of adaptive core points, N base To constrain the lower limit value, ρ local For the photon density along the orbit, A search Based on the adaptive neighborhood radius ε i η is the area of ​​a circle with radius η; η is the sparsity factor related to the estimated water depth.

[0023] In one preferred embodiment, the length of the local calculation window along the track is set according to the track resolution of the lidar data and the complexity of the lake topography.

[0024] In one preferred embodiment, dividing the lake into non-overlapping depth subspaces based on statistical distribution characteristics includes:

[0025] Calculate the probability density distribution of the continuous water depth distribution characteristics of the entire lake area, and determine the water depth delineation boundary based on the quantile of the cumulative distribution function of the probability density distribution.

[0026] In one preferred embodiment, the water quality parameters include:

[0027] Chlorophyll a, total phosphorus, total nitrogen, transparency, and permanganate index.

[0028] In one preferred embodiment, obtaining the indicators for evaluating the overall trophic status of a lake includes:

[0029] Based on the distribution of water quality parameters across the entire lake area, the comprehensive trophic state index is calculated using the comprehensive trophic state index method, and the trophic state level of the lake is classified according to the comprehensive trophic state index.

[0030] In one preferred embodiment, the comprehensive trophic state index method for calculating the comprehensive trophic state index includes weighting the trophic state index of the water quality parameters using the weights of the trophic state indices of the water quality parameters.

[0031] In one preferred embodiment, the lidar data is ICESat-2 data and the multispectral remote sensing data is Sentinel-2 data.

[0032] The technical solution of this invention has one or more of the following beneficial technical effects:

[0033] (1) By coupling the local terrain slope, the estimated water depth attenuation and the adaptive density clustering model of photon density along the track, the dynamic correction of water depth based on physical characteristics is realized, which solves the problem of high misclassification rate of lake surface photons and lake bottom photons in complex lake environment by the traditional fixed parameter DBSCAN model, and significantly improves the accuracy of lidar water depth measurement.

[0034] (2) The multispectral global water depth detection is calibrated by high-precision discrete water depth points of lidar, and the water depth subspace is adaptively divided based on the statistical distribution characteristics of water depth. The optimal water quality parameter detection model is constructed for different water depth subspaces, avoiding the sample imbalance problem caused by traditional equal depth distance manual division. It accurately adapts to the spectral response law of water quality parameters under different water depth conditions, and comprehensively improves the detection accuracy of various water quality parameters.

[0035] (3) By constructing a complete technical link of “LiDAR water depth extraction multispectral full-domain water depth detection adaptive stratified water quality detection comprehensive nutrient status evaluation”, the integrated detection of five core water quality indicators, namely chlorophyll a, total phosphorus, total nitrogen, transparency and permanganate index, and the accurate evaluation of lake eutrophication status have been realized. This has greatly improved the efficiency and coverage of lake water environment monitoring and provided reliable technical support for dynamic monitoring and refined management of lake water environment.

[0036] Furthermore, other beneficial effects of the present invention will be mentioned in the specific embodiments. Attached Figure Description

[0037] Figure 1 This is a flowchart of the main steps of one embodiment of the present invention;

[0038] Figure 2 This is a flowchart of the overall data processing process according to one embodiment of the present invention. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0040] To facilitate a clear description of the technical solutions in the embodiments of the present invention, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order.

[0041] The term "adaptive density clustering model" refers to a density-based clustering model (such as DBSCAN) that dynamically calculates clustering parameters (including adaptive neighborhood radius and adaptive minimum core number) for each photon point to be clustered, based on the physical characteristics of its local region, rather than using globally fixed parameter values. These physical characteristics include local terrain slope, estimated water depth attenuation (or estimated water depth), and along-track photon density. Through this adaptive process, the model can adapt to changes in water turbidity, topographic relief, and photon distribution density in different areas of the lake, thereby accurately identifying photon clusters on the lake surface and at the lake bottom.

[0042] The term "adaptive" refers to the fact that parameters or model structures in the data processing process are not preset to fixed values, but are automatically calculated and dynamically adjusted during operation based on the statistical or physical characteristics of the input data itself. Specifically, in the adaptive density clustering model of this invention, its "adaptiveness" is reflected in the fact that the clustering parameters (including neighborhood radius and minimum number of core points) are calculated in real time based on physical characteristics such as the terrain slope, estimated water depth, and along-track photon density of the local area where the photons to be clustered are located.

[0043] This invention discloses a stratified detection method for lake trophic status based on lidar and multispectral remote sensing. The method acquires high-precision discrete water depth points through lidar data, detects the water depth of the entire lake area by combining multispectral remote sensing data, and performs adaptive stratification based on the statistical distribution characteristics of water depth. Water quality parameter detection models are constructed for different water depth subspaces, and finally, the comprehensive trophic status of the lake is accurately evaluated.

[0044] Figure 1 This is a flowchart illustrating the main steps of one embodiment of the present invention. As shown in the figure, the method of the present invention includes four main steps:

[0045] Step S1: Extract lake depth data based on lidar data;

[0046] Step S2: Based on multispectral remote sensing data and the water depth point data, construct a water depth detection model and calculate the water depth of the entire lake area;

[0047] Step S3: Based on the water depth distribution characteristics of the entire lake area, construct a hierarchical remote sensing detection model for water quality parameters;

[0048] Step S4: Based on the stratified remote sensing detection model for water quality parameters, calculate the water quality parameters to obtain indicators for evaluating the overall trophic status of the lake. The following detailed explanation of each step is provided in conjunction with specific embodiments.

[0049] Figure 2 This is a flowchart illustrating the overall data processing process of one embodiment of the present invention. The flowchart clearly demonstrates the complete technical chain of the invention, including multi-source data input, core processing steps, and final output. Specifically, lidar data undergoes adaptive clustering and refraction correction to obtain high-precision discrete water depth point data. Multispectral remote sensing data is processed in two paths: one path extracts band reflectance, normalized difference water index (NDWI), and improved normalized difference water index (MNDWI) features, which are then combined with the water depth point data to obtain continuous water depth data for the entire lake area; the other path extracts band reflectance, ratio features, difference features, and normalized difference features to construct a full-spectrum feature set. Based on the lake's full-area water depth data, a water depth subspace is constructed. Layered detection models for each water depth subspace are trained using measured water quality data and the full-spectrum feature set. After accuracy verification and selection of the optimal model, the final output is the comprehensive trophic status evaluation result of the lake. The following detailed descriptions of each step are provided in conjunction with specific embodiments.

[0050] In this embodiment, a lake was selected as the study area. Ice, Cloud and land Elevation Satellite-2 (ICESat-2) data was used as the data source for water depth measurement, and Sentinel-2 multispectral satellite data was used as the remote sensing data source for water quality detection. At the same time, 30 sets of measured water quality data in the study area were collected for model training and validation.

[0051] First, step S1 is executed, which extracts lake depth data based on lidar data. This step further includes sub-steps S11, S12, and S13.

[0052] In sub-step S11, the lidar data is preprocessed. Specifically, the acquired ICESat-2 lidar data undergoes noise removal, elevation datum unification, strip stitching, and lake extent clipping operations sequentially. Noise removal aims to eliminate noise photons generated by atmospheric scattering, solar background light, etc., as well as land photons outside the lake. Elevation datum unification refers to converting the elevation values ​​of all photon points to the same vertical reference plane (e.g., Earth Gravity Model 2008 or Wusong Elevation System) to eliminate systematic biases between different data sources. Strip stitching is used to merge lidar data from multiple adjacent orbits into a complete dataset. Lake extent clipping uses pre-acquired lake vector boundaries to spatially clip the merged photon point cloud, thereby obtaining a dataset of pure water photons located only within the lake area. In this embodiment, after the above preprocessing, the total number of water photons within the lake area is 34044.

[0053] In sub-step S12, a Density-Based Spatial Clustering of Applications with Noise (DBSCAN) model coupled with multiple physical features is used to identify photon clusters on the lake surface and on the lake bottom. Traditional DBSCAN relies on a globally fixed neighborhood radius and minimum core point count. However, in a lake environment, the local terrain slope, water turbidity, and photon density along the trajectory vary significantly across different regions. Fixed parameters cannot adapt to these variations, easily leading to misclassification of photons on the lake surface and on the lake bottom. Therefore, this invention constructs a local calculation window along the trajectory direction for each water photon to be clustered. Within this window, the adaptive neighborhood radius and adaptive minimum core point count are dynamically calculated based on the local terrain slope around the water photon, the estimated water depth attenuation, and the photon density along the trajectory. In this embodiment, the trajectory length of the local calculation window is set to 50 meters, which can be adjusted according to the trajectory resolution of the lidar data and the complexity of the lake terrain.

[0054] The formula for calculating the adaptive neighborhood radius is as follows:

[0055] ,

[0056] Where, ε i Let ε be the adaptive neighborhood radius of the water photons to be clustered. base The base radius can be selected based on empirical values ​​according to the characteristics of different radar data. In this embodiment, ε base The value is 1.0 meter; k1 and k2 are adjustment coefficients, which can be selected based on empirical values ​​according to the turbidity of the water. In this embodiment, k1 is 0.5 and k2 is 0.3.

[0057] θ slope This refers to the local slope angle within a local calculation window, and its calculation method is as follows: In the formula σ z σ is the standard deviation of the elevation of all photon points within a local calculation window. x This represents the standard deviation of the orbital distances of all photons within the local window.

[0058] D est To estimate water depth, the calculation method is as follows: In the formula For the local water surface elevation within the local calculation window, z i The elevation of the water photons to be clustered;

[0059] D max The maximum water depth of the lake can be set based on historical lake data. In this embodiment, D... max The value is 30 meters.

[0060] In this embodiment, the method for determining the local water surface elevation is as follows: First, the approximate elevation range of the lake is obtained based on a global surface model. Then, within a local calculation window, a cluster of photon points with elevations within a 5-meter neighborhood of the lake's elevation is selected. The 95th percentile of the photon elevations within this cluster is taken as the local water surface elevation. If the estimated water depth D... est Greater than the maximum water depth D of the lake max If so, the photon to be clustered is discarded as noise.

[0061] The formula for calculating the adaptive minimum number of core points is as follows:

[0062] ,

[0063] Among them, MinPts i To achieve the minimum number of adaptive core points, N base To constrain the lower limit value, in this embodiment, N base The value is set to 3 to prevent overfitting caused by parameters that are too small. ρ is the floor function; local The average surface density of photons within a local calculation window, also known as the orbital photon density; A search To adapt the neighborhood radius ε i The area of ​​the circle is the radius; η is the sparsity factor, based on the estimated water depth D. est Using empirical values, in this embodiment, the estimated water depth D is taken as an example. est When the water depth is ≤10 meters, η is taken as 0.6. est When the distance is greater than 10 meters, η takes the value of 0.3.

[0064] For each water photon to be clustered, its adaptive neighborhood radius ε is calculated. i The total number of photon points within the core, if this total number is not less than the adaptive minimum core number MinPts. iIf a photon is found to be a core point, it is marked as a core point; otherwise, it is marked as a boundary point. Starting from each core point, photon points with achievable density within its adaptive neighborhood radius are recursively searched and included in the same cluster. After clustering, clusters with elevations close to the local water surface elevation and significantly higher density than the surrounding area are identified as lake surface photon clusters, while clusters located below the lake surface and continuously distributed along the track are identified as lake bottom photon clusters. In this embodiment, after adaptive DBSCAN clustering, a total of 11,640 effective photon points were obtained, including 11,348 lake surface photons and 292 lake bottom photons.

[0065] In this embodiment, the root mean square error (RMSE) of the water depth detection model based on the adaptive DBSCAN clustering model of the present invention is 1.16 meters, while the RMSE of the water depth detection model based on the traditional standard DBSCAN model is 2.28 meters.

[0066] In sub-step S13, lake depth data is calculated based on the identified lake surface photon clusters and lake bottom photon clusters. First, based on the lake surface photon clusters and the aforementioned local calculation windows, the median elevation of the lake surface photons within each window is calculated. Then, linear interpolation is performed along the track direction to generate a continuous and smooth water surface elevation curve along the track. In this embodiment, the interpolation interval is set to 1.0 meter to ensure that the curve is sufficiently smooth and undistorted. For each lake bottom photon, its apparent water depth relative to the water surface is first calculated, where the apparent water depth D is... i,0 The calculation formula is:

[0067] ,

[0068] Z surf,i Let z be the water surface elevation at the location corresponding to the lake bottom photon's position along its orbital water surface elevation curve. i The elevation of the photons at the bottom of the lake.

[0069] Because laser light refracts when propagating in water, the apparent water depth is not equal to the actual water depth, and refraction correction is required. The formula for calculating the actual water depth is:

[0070] , where ς is the refractive index of water. According to Snell's law and the refractive index of water (generally taken as 1.33), the theoretical value of the refractive index ς of water is approximately 0.75, and in this embodiment, the value is taken as 0.75. After the above calculation, the true water depth value corresponding to each lake bottom photon is obtained. These discrete water depth point data will be used as training samples for the multispectral water depth detection model in subsequent steps.

[0071] Next, step S2 is performed. Based on the multispectral remote sensing data and the water depth point data obtained in step S1, a water depth detection model is constructed to calculate the water depth of the entire lake area. This step further includes sub-steps S21, S22, and S23.

[0072] In sub-step S21, the multispectral remote sensing data is preprocessed. The acquired Sentinel-2 multispectral remote sensing data undergoes radiometric calibration, atmospheric correction, mosaicking, and lake extent cropping operations sequentially to obtain a multispectral remote sensing dataset covering the lake area. In this embodiment, the dataset contains reflectance data for six bands: blue (B), green (G), red (R), vegetation red-edge (VR), near-infrared (NIR), and shortwave infrared (SWIR).

[0073] In substep S22, the Normalized Difference Water Index (NDWI) and the Modified Normalized Difference Water Index (MNDWI) are calculated. NDWI and MNDWI are two commonly used water enhancement indices that can effectively suppress vegetation and soil information and highlight water features.

[0074] The formula for calculating NDWI is:

[0075] ,

[0076] The formula for calculating MNDWI is:

[0077] ,

[0078] Where ρ G ρ is the green band reflectance. NIR ρ is the reflectance in the near-infrared band. SWIR This refers to the reflectance in the shortwave infrared band. These two water indices will serve as important independent variables in the subsequent water depth detection model.

[0079] In sub-step S23, based on the lake depth data obtained in step S1, the multispectral remote sensing reflectance of each band obtained in sub-step S21, and the water index obtained in sub-step S22, a water depth detection model is constructed, with the reflectance of each band and the water index used as independent variables. The basic form of the water depth detection calculation model is as follows:

[0080]

[0081] Where D represents the water depth data (i.e., the dependent variable), and α i0 In this embodiment, α represents the initial coefficient of the i-th independent variable in the water depth detection model. i0 Both are 1; B iLet be the i-th independent variable in the water depth detection model. The independent variables include the reflectance of each band (e.g., blue band, green band, red band, vegetation red edge band, near-infrared band, short-wave infrared band) as well as NDWI and MNDWI, for a total of 8 independent variables (i.e., n=8). ε is the random error term.

[0082] Based on the lidar depth data obtained in step S1, regression analysis is performed using the least squares method to solve for the regression coefficients of each independent variable, thereby obtaining the trained depth detection model. The model is in the following form:

[0083] ,

[0084] in, α is the measured water depth value. i Let be the regression coefficient of the i-th independent variable, and n=8. By substituting the reflectance of each band and the water index of each multispectral pixel within the lake area into the model, the water depth value of each pixel in the entire lake area can be calculated, thereby obtaining the water depth distribution data of the entire lake area.

[0085] Then, step S3 is executed, which constructs a stratified remote sensing detection model for water quality parameters based on the lake-wide water depth distribution characteristics obtained in step S2. This step further includes sub-steps S31, S32, and S33.

[0086] In sub-step S31, the water depth classification threshold is adaptively determined based on the statistical distribution characteristics of the lake water depth data. Specifically, firstly, the probability density distribution of the global water depth data obtained in step S2 is calculated, which can be achieved using the kernel density estimation method. Then, the water depth classification boundary is adaptively determined based on the quantiles of the cumulative distribution function of the probability density distribution. Compared with the traditional manual classification method with equal depth intervals, this hierarchical approach based on statistical distribution ensures a relatively balanced number of samples within each water depth subspace and minimizes water depth variation within each subspace, which is beneficial for subsequent modeling. In this embodiment, the water depth values ​​corresponding to cumulative probability densities of 25%, 50%, and 75% are selected as classification boundaries, dividing the lake into four non-overlapping water depth subspaces with relatively balanced sample numbers.

[0087] It should be noted that the choice of quantiles can be adjusted according to the actual water depth distribution and sample size. For example, when the water depth distribution exhibits a multi-peak characteristic, the local minimum points of each probability density distribution can be selected as the dividing boundary.

[0088] In sub-step S32, the spectral feature set of the entire lake area is calculated, and the measured water quality parameters are randomly divided into training and validation sets. In this embodiment, five core water quality parameters are selected: chlorophyll a (Chla), total phosphorus (TP), total nitrogen (TN), transparency (Secchi Depth, SD), and permanganate index (CODMn). For each water quality parameter, 30 sets of measured sample data are collected and randomly divided into training set (21 sets) and validation set (9 sets) in a 7:3 ratio.

[0089] The method for calculating the spectral feature set of the entire lake area is as follows: Based on the reflectance of the six multispectral remote sensing bands obtained in sub-step S21, spectral ratio features, difference features, and normalized difference features are constructed. Specifically, for the six bands—blue band, green band, red band, vegetation red edge band, near-infrared band, and shortwave infrared band—the above three features are calculated by combining them in pairs.

[0090] Ratio Feature The calculation formula is:

[0091] ,

[0092] Difference feature Dif Feature The calculation formula is:

[0093] ,

[0094] Normalized difference feature NDif Feature The calculation formula is:

[0095] ,

[0096] Where, ρ i ρ j These are the reflectances of band i and band j, respectively, where i and j represent the six bands mentioned above. In this embodiment, 15 ratio features, 15 difference features, and 15 normalized difference features are calculated, for a total of 45 features. Adding these to the original six band reflectances, a total of 51 spectral features are obtained.

[0097] Then, for each water depth subspace, detection models with various combinations of spectral features are constructed. In this embodiment, five combination schemes are designed: Scheme 1 uses only single-band reflectance (6 features); Scheme 2 uses a combination of "single-band + ratio feature" (21 features); Scheme 3 uses a combination of "single-band + difference feature" (21 features); Scheme 4 uses a combination of "single-band + normalized difference feature" (21 features); and Scheme 5 uses all 51 spectral features. Based on the measured water quality parameter training set, a machine learning model is used to train a water quality parameter detection model for each feature combination scheme within each water depth subspace.

[0098] In sub-step S33, based on the measured water quality parameter validation set, accuracy analysis is performed on different water quality parameter detection models in each water depth subspace, and the detection model with the highest accuracy is selected. The accuracy analysis uses two indicators: the coefficient of determination (R²) and the root mean square error (RMSE). The coefficient of determination R²... i The formula for calculating ² is:

[0099] ,

[0100] Root Mean Square Error (RMSE) i The calculation formula is:

[0101] ,

[0102] in, The determination coefficient of the water quality parameter i detection model in this water depth subspace, where m is the number of samples of this water quality parameter in the validation set of measured water quality parameters, I test,j To verify the value of the j-th parameter of water quality parameter category i in the measured water quality parameter set, These are the measured values ​​of the water quality parameters for this sample. This is to validate the average value of water quality parameter category i in the measured water quality parameter set. RMSE i The root mean square error of the detection model for water quality parameter i in this water depth subspace is given.

[0103] Preferably, one can select The closest to 1 and RMSE i The smallest model is taken as the optimal detection model for the corresponding water quality parameter i within the water depth subspace.

[0104] The experimental results of this embodiment show that after adopting the hierarchical detection model, the highest R² values ​​for Chla, TP, TN, SD, and CODMn can reach 0.84, 0.84, 0.86, 0.94, and 0.78, respectively, and the lowest RMSE values ​​can reach 9.44, 0.02, 0.23, 7.87, and 0.45, respectively. In contrast, to verify the superiority of this invention, this embodiment also includes a comparative experiment: using a traditional unified water quality parameter detection model (i.e., without water depth stratification, using only a single model to cover the entire lake area), under the same training and validation set conditions, the R² values ​​for Chla, TP, TN, SD, and CODMn are 0.79, 0.77, 0.78, 0.82, and 0.74, respectively, and the RMSE values ​​are 13.87, 0.03, 0.32, 8.37, and 0.51, respectively. It is evident that the hierarchical modeling strategy of this invention significantly improves the detection accuracy of various water quality parameters, verifying the technical effectiveness of this invention.

[0105] In this embodiment, when using multispectral remote sensing data to detect lake water quality parameters and evaluate trophic status, the differences in the optical response of different water quality parameters to different water depths are fully considered. By constructing detection models based on different water depth subspaces in a hierarchical manner, the adaptability of the models is enhanced, and the accuracy of lake water quality parameter detection and trophic status evaluation is improved.

[0106] Finally, step S4 is executed, which calculates the distribution of water quality parameters across the entire lake area based on the stratified remote sensing model of water quality parameters obtained in step S3, and obtains indicators for evaluating the comprehensive trophic status of the lake. This step further includes sub-steps S41 and S42.

[0107] In sub-step S41, for each water depth subspace, the optimal detection model selected for each water quality parameter in sub-step S33 is used. The spectral features of all pixels in the subspace are substituted into the model to calculate the Chla, TP, TN, SD, and CODMn values ​​of each pixel, thereby obtaining the spatial distribution map of the five water quality parameters of the entire lake.

[0108] In sub-step S42, the five water quality parameters obtained in sub-step S41 are used to conduct a comprehensive trophic status assessment of the lake. This embodiment uses the Tropical Level Index (TLI) method to conduct the comprehensive trophic status assessment of the lake. The formula for calculating the Tropical Level Index is:

[0109] ,

[0110] Where TLI(Σ) is the comprehensive trophic state index, n is the number of water quality parameters (n=5 in this embodiment), and W pis the weight of the trophic level index for the p-th water quality parameter, and TLI(p) is the trophic level index for the p-th water quality parameter.

[0111] The weight is determined by the correlation coefficient method. Chlorophyll a (Chla) is taken as the reference parameter, the actual measured correlation coefficients between the other four parameters (TP, TN, SD, CODMn) and Chla are calculated, and then the weight of each parameter is calculated according to the following formula:

[0112] ,

[0113] wherein R ij is the correlation coefficient between the j-th parameter and the reference parameter i. In this embodiment, the actually measured water quality parameter Chla is taken as the reference parameter, and the summation symbol traverses all 5 parameters.

[0114] The trophic level index of each water quality parameter is obtained respectively through the following calculation formulas:

[0115] TLI(Chla)=10×(2.5+1.086×ln(Chla));

[0116] TLI(TP)=10×(9.436+1.624×ln(TP));

[0117] TLI(TN)=10×(5.453+1.694×ln(TN));

[0118] TLI(SD)=10×(5.118-1.94×ln(SD));

[0119] TLI(CODMn)=10×(0.109+2.661×ln(CODMn));

[0120] In the above formulas, ln represents the natural logarithm operation.

[0121] According to the calculated comprehensive trophic level index TLI(Σ), the trophic level grades of the lake are divided according to the following criteria: when TLI(Σ)≤30, it is oligotrophic; when 30<TLI(Σ)≤50, it is mesotrophic; when 50<TLI(Σ)≤60, it is light eutrophic; when 60<TLI(Σ)≤70, it is moderate eutrophic; when TLI(Σ)>70, it is severe eutrophic.

[0122] In summary, the lake trophic status stratification detection method based on lidar and multispectral remote sensing provided in this invention achieves dynamic water depth correction based on physical characteristics by introducing an adaptive clustering model that couples local topographic slope, estimated water depth attenuation, and along-track photon density, significantly improving the accuracy of lidar water depth measurement. Furthermore, based on the statistical distribution characteristics of lake water depth, adaptive stratification is performed to construct optimal water quality parameter detection models for different water depth subspaces, fully considering the influence of water depth differences on optical response laws, effectively improving the detection accuracy of multiple water quality parameters such as chlorophyll a, total phosphorus, total nitrogen, transparency, and permanganate index. Finally, a comprehensive trophic status index method is used to accurately evaluate the eutrophication state of lakes. This method features high automation, a clear physical mechanism, and excellent detection accuracy, and has significant scientific and practical value for the dynamic monitoring and management of lake trophic status.

[0123] The specific implementation methods of the above modules are consistent with the corresponding steps in the aforementioned method embodiments, and will not be repeated here.

[0124] To better illustrate the present invention, numerous specific details have been provided in the detailed embodiments described above. Those skilled in the art should understand that the present invention can be practiced even without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of the present invention.

[0125] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for stratified detection of lake trophic status based on lidar and multispectral remote sensing, characterized in that, include: Acquire lidar data and use an adaptive density clustering model that couples local terrain slope, estimated water depth and photon density along the track to identify lake surface photon clusters and lake bottom photon clusters in the lidar data, and obtain water depth point data after refraction correction. Based on the aforementioned water depth point data and multispectral remote sensing data, the continuous water depth distribution characteristics of the entire lake area are obtained. Based on the continuous water depth distribution characteristics of the entire lake area, the lake is divided into non-overlapping water depth subspaces based on statistical distribution characteristics. For each of the aforementioned water depth subspaces, a water quality parameter detection model is constructed and trained, specifically including: for each of the aforementioned water depth subspaces, based on the original band reflectance of the multispectral remote sensing data, constructing spectral ratio features, difference features, and normalized difference features; and using measured water quality data, selecting the feature combination with the highest accuracy from a full-spectrum feature set containing the original band reflectance, the spectral ratio features, the difference features, and the normalized difference features to train the water quality parameter detection model; and Based on the measured water quality data, the water quality parameter detection model with the highest accuracy is selected as the optimal detection model for this water depth subspace; and Based on the optimal detection model for each water depth subspace, the distribution of water quality parameters across the entire lake area is calculated, and indicators for evaluating the comprehensive trophic status of the lake are obtained. The adaptive density clustering model dynamically calculates the adaptive neighborhood radius and the adaptive minimum core point number for each photon point to be clustered, based on the local terrain slope, estimated water depth, and photon density along the track within the local calculation window where the photon point is located. Clustering is then performed based on the adaptive neighborhood radius and the adaptive minimum core point number. The formula for calculating the adaptive neighborhood radius is as follows: ， Where, ε i For adaptive neighborhood radius, ε base The base radius, k1 and k2 are adjustment coefficients, and θ slope D represents the local slope angle within the local calculation window. est To estimate the water depth, D max This represents the maximum depth of the lake.

2. The method for stratified detection of lake trophic status based on lidar and multispectral remote sensing according to claim 1, characterized in that, The formula for calculating the adaptive minimum number of core points is as follows: ， Among them, MinPts i To achieve the minimum number of adaptive core points, N base To constrain the lower limit value, ρ local For the photon density along the orbit, A search Based on the adaptive neighborhood radius ε i η is the area of ​​a circle with radius η; η is the sparsity factor related to the estimated water depth.

3. The method for stratified detection of lake trophic status based on lidar and multispectral remote sensing according to claim 2, characterized in that: The length of the local calculation window along the track is set according to the track resolution of the lidar data and the complexity of the lake topography.

4. The method for stratified detection of lake trophic status based on lidar and multispectral remote sensing according to claim 1, characterized in that, The method of dividing lakes into non-overlapping depth subspaces based on statistical distribution characteristics includes: Calculate the probability density distribution of the continuous water depth distribution characteristics of the entire lake area, and determine the water depth delineation boundary based on the quantile of the cumulative distribution function of the probability density distribution.

5. The method for stratified detection of lake trophic status based on lidar and multispectral remote sensing according to claim 1, characterized in that, The water quality parameters include: Chlorophyll a, total phosphorus, total nitrogen, transparency, and permanganate index.

6. The method for stratified detection of lake trophic status based on lidar and multispectral remote sensing according to claim 1, characterized in that, The indicators used to evaluate the overall trophic status of lakes include: Based on the distribution of water quality parameters across the entire lake area, the comprehensive trophic state index is calculated using the comprehensive trophic state index method, and the trophic state level of the lake is classified according to the comprehensive trophic state index.

7. The method for stratified detection of lake trophic status based on lidar and multispectral remote sensing according to claim 6, characterized in that: The comprehensive trophic state index method calculates the comprehensive trophic state index by weighting the trophic state index of water quality parameters using the weights of the trophic state indices of the water quality parameters.

8. The method for stratified detection of lake trophic status based on lidar and multispectral remote sensing according to claim 1, characterized in that: The lidar data is ICESat-2 data, and the multispectral remote sensing data is Sentinel-2 data.

Images