Remote sensing synchronous inversion method for lake water quality and terrain based on watershed-shoreline cooperative constraint

Abstract

The invention relates to a remote sensing synchronous inversion method for lake water quality and terrain based on basin-shoreline cooperative constraint, which comprises the following steps of: training an XGBoost model by taking lake meteorological characteristics, coastal and offshore terrain characteristics, Landsat-8 wave band reflectivity data and spectral index data as input and taking synchronous actually-measured underwater terrain data and actually-measured water quality data as output; and selecting the optimal model as an inversion model of the underwater topographic data and the water quality data, and inverting the spatial-temporal distribution of the underwater topographic data and the water quality data of the lake by using the estimation model. According to the invention, through fusion of watershed terrain, meteorological elements and lake shoreline shape features, limitation that traditional remote sensing inversion only pays attention to lake internal spectral response is broken through, and remote sensing synchronous estimation of lake water quality and terrain under conditions of insufficient data information or difficult field investigation (such as plateau lakes) is supported; and reliable technical support and reference basis are provided for collaborative management of lake water environment and water resources.

Description

Technical Field

[0001] The present invention belongs to the field of environmental remote sensing, and specifically relates to a remote sensing synchronous inversion method for lake water quality and terrain with basin-shoreline collaborative constraints. Background Art

[0002] Remote sensing technology has become an important tool for monitoring lake water environments, enabling high-precision quantitative extraction of key parameters such as chlorophyll a, suspended solids, colored dissolved organic matter (CDOM), and transparency. Early studies primarily established empirical relationships between remote sensing spectra and the concentrations of pigments in lake substances, enabling the quantitative inversion of specific lake water quality parameters. In recent years, with the rapid development of intelligent algorithms such as machine learning, numerous studies have used algorithms such as random forests and neural networks to train spectral bands or combinations of different bands, significantly improving the inversion accuracy and generalization capabilities of these models and promoting the widespread application of these technologies. However, existing research indicates that the mechanisms driving the spatial distribution of lake substance concentrations are extremely complex. In shallow lake ecosystems, wind and wave disturbances act as a significant physical driving force, altering the mixing state of the water and promoting significant spatial and temporal migration of algal populations. Unique hydrodynamic conditions in special topographical areas, such as bays, create relatively stable algal aggregations. It is noteworthy that changes in lake water environments are essentially the result of the interaction between human activities and meteorological factors within the watershed. Human activities, such as land use change and point source emissions, directly alter nutrient input fluxes, while meteorological factors, such as precipitation and temperature, jointly regulate lake nutrient concentrations and their spatial distribution by influencing water exchange processes. Therefore, by incorporating lake-wide spectral data into training, considering the interaction between the watershed and lake system, and combining lakeshore characteristics with climatic factors to improve model generalization, new approaches can be explored for constructing comprehensive models with both mechanistic support and inversion performance.

[0003] In the field of lake water volume monitoring, the scarcity of measured data leads to significant uncertainty in total lake water volume estimates. The core of lake water volume surveys lies in measuring lake depth and underwater topography. Existing research focuses on two technical approaches: direct measurement and spatial inference. The former focuses on accurately extracting underwater elevation information using multi-platform sensors such as ship-borne sonar and drone-mounted lidar. The latter aims to construct spatial inference models based on surrounding topographic features to infer underwater topography in areas without field measurements. Research on lake depth estimation using optical imagery and satellite altimetry data remains limited. While remote sensing satellites have proven effective for bathymetric mapping in intermittently flooded areas, obtaining topographic information in permanently flooded areas remains challenging. To overcome these challenges, several spatial prediction and modeling methods have been proposed to predict underwater topography by exploiting the exposed topography surrounding lakes. The relationship between topographic variables and lake depth (including lake depth and volume) is well-studied, providing support for developing geostatistical models for estimating lake depth and volume at regional and even global scales. However, these estimates are generally acceptable in large-scale studies because the underestimation and overestimation biases offset each other. However, the uncertainties and potential biases at the lake-specific or local scale cannot be ignored. Therefore, a specific and effective spatial modeling approach is needed to enhance the responsiveness of spatial inference models both above water (watershed topography, lake shoreline) and below water (water quality parameters), providing an optimized solution for the set of topographic variables.

[0004] In summary, conducting simultaneous remote sensing monitoring of lake water environment and water resources and constructing a simultaneous inversion model of lake water quality and topography are not only the scientific basis for accurately assessing lake ecological health and supporting river basin water environment management decisions, but also an urgent need to promote the sustainable use of water resources and the construction of ecological civilization in river basins. This invention integrates remote sensing spectral inversion, river basin surface information, and machine learning algorithms to break through the technical bottlenecks of traditional single-element monitoring, constructing a water quality-topography simultaneous inversion framework, and providing an integrated solution for multi-dimensional monitoring of lake ecosystems. Summary of the Invention

[0005] The present invention aims to provide a method for simultaneous remote sensing inversion of lake water quality and topography, using coordinated basin-shoreline constraints. The principles, processes, and results of this method support simultaneous remote sensing estimation of lake water quality and quantity in conditions where data is insufficient or field surveys are difficult (e.g., plateau lakes). This method can also provide a foundation for Landsat satellites to simultaneously acquire lake water environment and water resource information.

[0006] To achieve the above object, the present invention adopts the following scheme: A method for simultaneous remote sensing inversion of lake water quality and topography with basin-shoreline collaborative constraints, the method comprising: The lake boundary is divided into several segments; for each segment, the slope, aspect, and elevation data of the offshore distance sequence are extracted based on the DEM data by extending the segment median perpendicular to the land; and the ratio of the segment shoreline length to the segment straight line length is calculated for each segment of the lake boundary as the shoreline complexity, which is combined with the slope, aspect, and elevation data to form the lake's coastal and offshore terrain feature data; Obtaining lake meteorological characteristic data and band reflectance data of Landsat-8 remote sensing data covering the lake; and calculating the algae biomass index ABI and the water color index FUI based on the band reflectance data; The XGBoost model is trained using the terrain features, meteorological features, ABI, FUI, and reflectance data of green, red, and near-infrared bands as input, and synchronized measured underwater terrain data and measured water quality data as output; The optimal model is selected as the inversion model of underwater topography data and water quality data, and the spatiotemporal distribution of the lake underwater topography data and water quality data is inverted using the inversion model.

[0007] In some embodiments of the present invention, after extracting the lake water range based on the band reflectance data, the lake boundary is obtained based on the extracted lake water range.

[0008] In some embodiments of the present invention, the boundary of the lake is divided into several segments with equal distances.

[0009] In some embodiments of the present invention, for each segmented perpendicular line extending landward, slope, aspect and elevation data are extracted at equal intervals.

[0010] In some embodiments of the present invention, the method further includes determining the interval distance: setting different interval distances for the segmented perpendicular midlines, and selecting the interval distance that optimizes the model accuracy as a sampling parameter for the offshore terrain data.

[0011] In some embodiments of the present invention, lakes used for model training include lakes of different shapes, different shoreline complexity, and different water quality data coverage.

[0012] In some embodiments of the present invention, the meteorological characteristics are daily mean data obtained based on the ERA5 reanalysis dataset, including temperature, radiation, precipitation and wind speed.

[0013] In some embodiments of the present invention, the method further includes, after obtaining the inverted lake underwater topography data, estimating the water volume of the lake in combination with the lake water level and area.

[0014] In some embodiments of the present invention, the Landsat-8 remote sensing data uses the L2 surface reflectance product of the Landsat-8 OLI sensor; the surface reflectance is divided by π to obtain the water reflectance. R rs (λ), and R rs (λ) The dataset is subjected to outlier removal: the reflectance value of the band is less than 0, or the reflectance of the green band is less than 0.02 sr -1 , or any visible light band reflectivity greater than 0.50 sr -1 The pixels are masked.

[0015] In some embodiments of the present invention, during model training, a ten-fold cross-validation method is used to evaluate model performance in combination with the average root mean square error and average deviation of the prediction results, and the model with the best comprehensive accuracy is selected as the final inversion model.

[0016] Based on a machine learning algorithm that uses collaborative constraints on watersheds and shorelines, this paper develops a joint lake water quality and topography inversion model suitable for the Landsat-8 OLI sensor. Landsat-8 is a wide-band satellite with fewer spectral bands, capturing less spectral information than the more commonly used MODIS ocean color satellite. Therefore, it is less commonly used in lake color inversion. During model training, the present invention selects lake samples of varying shapes and water quality ranges to improve the model's applicability. In addition to the visible and near-infrared bands, the model also incorporates spectral indices indicating algal biomass and water color when selecting input features. These remote sensing spectral indices are correlated with nutrient levels and transparency in the water, accelerating model convergence. The input terrain feature data includes coastal and offshore data, which enables spatial deduction of underwater terrain. Further, based on the distribution of water quality parameters, the correlation between terrain data and meteorological elements (such as extreme precipitation and wind erosion may change the shape of the lake, thereby affecting the topography near the lakeshore or lakeside zone), specific meteorological data is selected as input. Based on the selected lake sample characteristics and 4-dimensional lake characteristics, a high-precision inversion model is trained.

[0017] By integrating basin topography, meteorological elements, and lake shoreline shape characteristics, this invention breaks through the limitation of traditional remote sensing inversion that only focuses on the internal spectrum of lakes. It supports the simultaneous remote sensing estimation of lake water quality and quantity under conditions of insufficient data or difficult field investigations (such as plateau lakes). The principles, processes, and results of the method can provide a reference basis for the simultaneous estimation of key elements of lake water environment and lake water resource reserves by Landsat series satellites.

[0018] It should be understood that all combinations of the foregoing concepts and the additional concepts described in more detail below, as long as such concepts are not mutually inconsistent, can be considered part of the inventive subject matter of this disclosure. In addition, all combinations of the claimed subject matter are considered part of the inventive subject matter of this disclosure.

[0019] The foregoing and other aspects, embodiments, and features of the present invention will be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the present invention, such as features and / or beneficial effects of the exemplary embodiments, will become apparent from the following description or through practice of specific embodiments according to the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings are not intended to be drawn to scale. In the accompanying drawings, each identical or nearly identical component shown in various figures may be represented by the same reference numeral. For clarity, not every component is labeled in every figure, wherein: Figure 1 The spatial distribution of the measured underwater topography of the sampled lakes (a) Lake A and (b) Lake B.

[0021] Figure 2 It is the principle of deducing underwater topography of lakes.

[0022] Figure 3 This is the flow chart of the XGB-LWQQ algorithm.

[0023] Figure 4 It is the influence of surface data extraction interval on algorithm accuracy.

[0024] Figure 5 The prediction accuracy of the XGB-LWQQ algorithm based on (a) Chla, (b) SD, and (c) underwater terrain data of the training set; the prediction accuracy of the XGB-LWQQ algorithm based on (d) Chla, (e) SD, and (f) underwater terrain data of the validation set.

[0025] Figure 6 It is the variation of underwater terrain inversion accuracy with offshore distance based on the XGB-LWQQ algorithm.

[0026] Figure 7 This is a comparison of the accuracy of the XGBoost algorithm with the DNN, SVR, KNN, MLR and RF algorithms.

[0027] Figure 8 Landsat-8 OLI images, average chlorophyll-a concentration, transparency spatial distribution, and underwater topography simulation results of Lake A and Lake B. DETAILED DESCRIPTION

[0028] In order to better understand the technical content of the present invention, specific embodiments are given and described below with reference to the accompanying drawings.

[0029] Various aspects of the present invention are described in this disclosure with reference to the accompanying drawings, in which a number of illustrative embodiments are shown. The embodiments of the present disclosure are not necessarily intended to include all aspects of the present invention. It should be understood that the various concepts and embodiments introduced above, as well as those described in more detail below, can be implemented in any of many ways, and that the concepts and embodiments disclosed herein are not limited to any implementation. In addition, some aspects of the present disclosure may be used alone or in any appropriate combination with other aspects of the present disclosure.

[0030] Example 1 This embodiment takes Lake A and Lake B in a certain plateau province as examples to further describe the technical solution of the present invention.

[0031] The water quality data of the two lakes selected in this example covers a wide range and can well represent the optical properties of the water bodies of plateau lakes. At the same time, the shapes of the two lakes are both narrow and long, elliptical, and irregularly concave, ensuring that the developed model has good applicability.

[0032] The models of the present invention can be trained independently (one output per model) or jointly (using XGBoost multi-output regression). In the embodiment, independent training is used to build the model.

[0033] The method of the present invention is based on the measured chlorophyll a data, transparency, lake underwater topography data and synchronized Landsat-8 OLI surface reflectance data of the study area, and obtains the off-water reflectance through simple correction. R rs ; Using three bands of visible light R rs , 1 near-infrared band and 2 remote sensing spectral indices as input, and simultaneously input 4 watershed terrain characteristics (slope, aspect, elevation, shoreline complexity) and 4 meteorological characteristics (temperature, radiation, precipitation, wind speed), and used the machine learning method XGBoost to develop a lake water quality and terrain inversion algorithm (XGB-LWQQ) suitable for Landsat-8; then, the algorithm was applied to some lake cases to evaluate its performance; finally, the algorithm was applied to long-term image series to obtain the changing trends of lake water quality and topography from 2013 to 2020.

[0034] As an exemplary description, the implementation of the above method is described in detail below with reference to the accompanying drawings.

[0035] Step 1: Based on Landsat-8 remote sensing product data, obtain a surface reflectance (SR) dataset of four bands; collect lake underwater topography data, as well as measured chlorophyll a concentration and transparency data of the lake during the Landsat-8 synchronous transit time.

[0036] The L2 level remote sensing product of the Landsat-8 OLI sensor is used as the data source. To correct the influence of the solar zenith angle, the SR is divided by π to obtain the water reflectance. R rs (λ). At the same time, the reflectivity value of the band is less than 0, or the reflectivity of the green band is less than 0.02 sr -1 , or the reflectivity in the visible light band is greater than 0.50 sr -1 The outliers are removed to obtain the results of the areas covered by water bodies only.

[0037] The underwater topographic data of the lake were collected using a multi-beam echosounder, and the acquisition time, coordinate system and vertical datum parameters of the echosounder data were recorded simultaneously.

[0038] Step 2: Data and algorithm structure used in the development of the XGB-LWQQ algorithm; Based on the existing lake water quality parameter sampling information, the following criteria were used to screen the data: (1) combining field records and satellite images to remove the sample points covered by clouds and solar flares; (2) using a ±3h time window to match the dataset; (3) retaining the 3×3 window of each sample point. R rs (λ) The coefficient of variation is within 10%. According to the above standards, a total of 211 groups of water quality sample data were obtained.

[0039] The measured lake water quality data were randomly divided into two categories, 150 of which were used for modeling and 61 for verification. The concentrations of substances in the two lakes were significantly different. The average Chl-a of Lake B was 1.09±0.16μg / L, which is an oligotrophic lake with an average transparency greater than 10m; the average Chl-a of Lake A was 41.45±21.99μg / L, which is a eutrophic lake with a transparency of 1.32±0.36m. In addition, the water area of ​​Lake A is long and narrow ( Figure 1 In the middle (a), the lakeshore is relatively straight; the waters of Lake B are butterfly-shaped ( Figure 1 The lakeshore is complex and curved, with many bays.

[0040] The normalized difference water index (NDWI) was constructed based on the green band (561nm) and near-infrared band (865nm) of Landsat-8 OLI to extract the lake water area.

[0041] The OTSU algorithm was used to determine the water-land segmentation threshold, extract the lake water area, and obtain the synchronous lake boundary during the image transit period; the lake boundary was divided into equidistant segments, and the fractal dimension of each segment was calculated based on fractal theory to indicate the complexity of the lake shoreline. The slope and slope direction were calculated using the DEM data within the basin; for each lake shoreline segment, the slope was extended to the land along its perpendicular line and the slope was extracted at equal intervals along the offshore distance sequence ( tan ( α )), Slope (angle α , normalized) and elevation values.

[0042] In this example, the boundary between Lake A and Lake B is divided into 500 equal segments, and the boundary length is calculated using the shapely.geometry package of Python. Based on fractal theory, the length of the curve between the points (the length of the shoreline of this segment) is calculated. L shoreline ) and the straight line length (the distance between the first and last endpoints of the coastline D shoreline ) ratio, which represents the complexity of the lake shoreline. The closer the ratio is to 1, the straighter the shoreline is, and vice versa, the more tortuous it is. Figure 1 ).

[0043] The digital elevation model (DEM) data of the watershed surface with a spatial resolution of 30 meters was obtained from the Geospatial Data Cloud (https: / / www.gscloud.cn / ), and the slope and slope direction were calculated using Arcmap. The offshore surface elevation was taken every 100 meters along the median perpendicular line of the segmented coastline ( h 1… h k ), slope and aspect values ​​(such as Figure 2 As shown, the present invention uses offshore surface elevation to infer underwater topography h '1… h ' n ), together with the segmented coastline complexity, are used as 4D terrain information features for machine learning algorithm training.

[0044] Studies have shown that the algal biomass index (ABI) has a good response to the nutrient status index based on Chl-a. At the same time, the water color index (FUI) has been found to have a power function decay relationship with transparency, and the determination coefficient R 2 As high as 0.95. Since lake transparency is mainly affected by the chromaticity angle α, the FUI index is introduced as an input feature.

[0045] Considering the close correlation between meteorological factors and the spatial distribution of lake water quality parameters, and the fact that extreme precipitation and wind erosion can alter lake shape, significantly affecting the topography of the nearshore or lakeshore zone, this example uses meteorological elements from the ERA5 global reanalysis dataset, including surface temperature, solar radiation, wind speed, and precipitation data. This dataset has a spatial resolution of 0.25° × 0.25° and a temporal resolution of 1 hour. Based on the smallest watershed unit where the lake is located, the daily mean values ​​of watershed temperature, solar radiation, wind speed, and precipitation during the same period of satellite overpass are calculated. These four-dimensional meteorological element features serve as input for model training.

[0046] The XGB-LWQQ method process is as follows Figure 3 As shown in the figure, terrain information features, remote sensing spectral features, and meteorological element features are input into the model for training. The XGB-LWQQ method is implemented using the Python XGBoost package. Parameters are tuned during model training. These include the learning rate, maximum tree depth, minimum leaf node sample weights, maximum leaf node number (Tree num), random sampling ratio for training samples, random sampling ratio for features, and the regularization parameter. A grid search method is used to optimize parameter combinations. Target parameters include the learning rate, tree depth, and minimum leaf node weights. Parameter combinations that meet accuracy requirements and minimize the model complexity metric (tree depth × number of leaf nodes) are prioritized to reduce model complexity and enhance generalization. Ultimately, the optimal model structure parameters are determined.

[0047] In this embodiment, the input parameters are optimized. A variety of XGBoost machine learning algorithms are constructed by permutation and combination of the input parameters. The best accuracy is selected as the final model input. Tables 1 to 3 are the model inversion accuracy of each output parameter under different input parameter combinations. Due to the large number of parameter combinations, only R is retained for the water quality parameter inversion in the table. 2 Models with values ​​≥ 0.7 are used as reference for comparison, and R is retained for underwater terrain inversion. 2 ≥0.5. During input parameter screening, 200 samples were selected from the total sample data for model training. Tables 1-3 show the training set accuracy.

[0048] Table 1. Chla inversion accuracy evaluation results for different input parameters

[0049] In the table, N refers to the sample size.

[0050] Table 2 Transparency inversion accuracy evaluation results corresponding to different input parameters

[0051] Table 3 Inversion accuracy evaluation results of underwater terrain corresponding to different input parameters

[0052] Based on the model accuracy evaluation results, Landsat-8 OLI was finally selected. R rs (561), R rs (655), R rs (865), algal biomass index (ABI) and water color index (FUI), a total of 5-dimensional remote sensing spectral features, combined with meteorological characteristics and coastal and offshore topographic characteristics are used as inputs for model training.

[0053] In this embodiment, the model parameters obtained by training through the above-mentioned method principle are: learning rate is 0.05, the maximum depth of the tree is 4 layers, the resampling rate is 0.8, the maximum number of leaf nodes is 100, the regularization parameter adopts L2 regularization coefficient 0.01, the maximum number of iterations is initially 1000 times, and the random sampling ratio of training samples is 7:3.

[0054] This example also analyzes the impact of different sampling intervals on the median vertical line on model accuracy. The sampling intervals of the lakeshore median vertical line are set to 50m, 100m, 200m, and 500m, respectively, and four XGBoost machine learning models are constructed. At the same time, in order to ensure the stability of the model and its independence from the training set, the ten-fold cross-validation method is used to evaluate the model performance (average RMSE and MAPE) using 10 randomly selected training and validation sets. The results are shown in the figure below. Figure 4 As shown, it can be seen that the model accuracy and sampling interval distance are not linearly related. The accuracy of the sampling interval of 100m is better than that of 50m, 200m and 500m. Therefore, this embodiment selects the sampling interval with the highest accuracy (100m) as the surface data extraction interval.

[0055] Under the optimal input parameter combination and sampling accuracy, the XGB-LWQQ algorithm shows high accuracy on both training and validation datasets ( Figure 5 ), most of the data points in the training set are evenly distributed near the 1:1 line, and the accuracy of the validation set is slightly lower than that of the training set. The accuracy statistics of the water quality inversion model show that the average accuracy of the Chla and SD training sets in the cross validation is R 2 The results of Chla and SD inversion were 0.98 and 0.96, RMSE was 6.21 μg / L and 0.48 m, MAPE was 15.78% and 32.43% (N = 182). R 2The results were 0.86 and 0.88, RMSE were 10.72 μg / L and 0.92 m, and MAPE were 28.96% and 45.97% (N = 94). R 2 is 0.96, RMSE and MAPE are 1.56m and 0.08% (N = 350), respectively; the average R 2 The RMSE and MAPE were 5.26m and 0.27% (N = 150), respectively. Multiple rounds of validation indicate that the water quality and terrain inversion models are less dependent on the training set and have good anti-interference performance, indicating that the inversion models have good generalization capabilities.

[0056] However, due to the diverse causes of lake formation, complex geological structures, and limited spectral signal penetration, the inversion accuracy of using terrestrial surface features to infer underwater topography is relatively high in lakes with regular shapes and regular changes in lake basin topography (RMSE of Lake A is 5.44m), while the inversion accuracy is relatively low when the lake shape is irregular or the water depth is deep (RMSE of Lake B is 8.30m).

[0057] This example further extracts underwater terrain inversion values ​​at different offshore distances and compares them with the measured underwater terrain results. As the offshore distance increases, the RMSE increases from 5.15m to 7.24m, the MAPE increases from 0.26% to 0.32%, and the error increases by 0.06% per kilometer ( Figure 6 ). Therefore, the accuracy of underwater terrain inversion in areas far from the lakeshore is slightly lower than that of the terrain inversion results close to the lakeshore.

[0058] This example further compares the accuracy results of the XGBoost algorithm with several other common intelligent algorithms ( Figure 7 ), including deep neural network (DNN), support vector machine (SVR), K-nearest neighbor (KNN), multiple linear regression (MLR) and random forest (RF). Limited by the number of training model samples, the accuracy of the deep learning DNN model is relatively poor, the prediction results are overfitted, and the RMSE is close to 50m. The accuracy of SVR and KNN is significantly better than that of the DNN algorithm, and the scattered results are relatively evenly distributed on both sides of the 1:1 line. The RMSE and MAPE of the multiple linear regression, random forest and XGBoost algorithms are relatively close, but the bias (BIAS) of the XGBoost algorithm is relatively smaller, only -0.51. Therefore, the performance of the XGB-LWQQ algorithm of the present invention is better than other mainstream intelligent algorithms.

[0059] Step 3: Apply the lake water quality and topography inversion algorithm to the lake remote sensing data to obtain the spatial distribution of lake chlorophyll a, transparency, and underwater topography.

[0060] Figure 8 The XGB-LWQQ inversion results for Lake A and Lake B are presented. The spatial distribution of chlorophyll a in the lakes obtained using the XGB-LWQQ algorithm is consistent with that in the RGB imagery. For example, Lake A has high chlorophyll a concentration in the eastern part and low concentration in the western part. Furthermore, the spatial distribution of transparency shows an opposite trend to the chlorophyll a concentration distribution: areas with abundant algae and high chlorophyll a concentration have relatively low transparency, while clearer, more transparent waters have relatively low chlorophyll a concentrations. The spatial distribution of underwater topography indicates that the elevation of Lakes A and B decreases from the shore to the center. The underwater topography on the southern shore of Lake A changes rapidly, while the topography on the eastern bay of Lake B varies significantly, with relatively high topography. These findings are consistent with the measured spatial distribution of underwater topography in lakes, demonstrating good simulation accuracy.

[0061] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Those skilled in the art will appreciate that various modifications and variations can be made without departing from the spirit and scope of the present invention.

Claims

1. A method for simultaneous remote sensing inversion of lake water quality and topography with basin-shoreline collaborative constraints, characterized by: The method comprises: The lake boundary is divided into several segments; for each segment, the slope, aspect, and elevation data of the offshore distance sequence are extracted based on the DEM data by extending the segment median perpendicular to the land; and the ratio of the segment shoreline length to the segment straight line length is calculated for each segment of the lake boundary as the shoreline complexity, which is combined with the slope, aspect, and elevation data to form the lake's coastal and offshore terrain feature data; Obtaining lake meteorological characteristic data and band reflectance data of Landsat-8 remote sensing data covering the lake; and calculating the algae biomass index ABI and the water color index FUI based on the band reflectance data; The XGBoost model is trained using the terrain features, meteorological features, ABI, FUI, and reflectance data of green, red, and near-infrared bands as input, and synchronized measured underwater terrain data and measured water quality data as output; The optimal model is selected as the inversion model of underwater topography data and water quality data, and the spatiotemporal distribution of the lake underwater topography data and water quality data is inverted using the inversion model.

2. The method according to claim 1, characterized in that After extracting the lake water range based on the band reflectance data, the lake boundary is obtained based on the extracted lake water range.

3. The method according to claim 1, characterized in that Divide the lake boundary into several segments at equal distances.

4. The method according to claim 1, wherein For each segmented perpendicular line extending landward, the slope, aspect and elevation data are extracted at equal intervals.

5. The method according to claim 4, characterized in that The method further includes determining the interval distance: setting different interval distances for the segmented perpendicular midlines, and selecting the interval distance that optimizes the model accuracy as a sampling parameter for the offshore terrain data.

6. The method according to claim 1, characterized in that The lakes used for model training include lakes of different shapes, different shoreline complexity, and different water quality data coverage.

7. The method according to claim 1, characterized in that The meteorological characteristics are daily mean data obtained based on the ERA5 reanalysis dataset, including temperature, radiation, precipitation, and wind speed.

8. The method according to claim 1, characterized in that The method also includes estimating the water volume of the lake by combining the water level and area of ​​the lake after obtaining the inverted lake underwater topography data.

9. The method according to claim 1, wherein The Landsat-8 remote sensing data uses the L2 surface reflectance product of the Landsat-8OLI sensor. The surface reflectance is divided by π to obtain the water reflectance. R rs (λ), and R rs (λ) The dataset is subjected to outlier removal: the reflectance value of the band is less than 0, or the reflectance of the green band is less than 0.02 sr -1 , or any visible light band reflectivity greater than 0.50 sr -1 The pixels are masked.

10. The method according to claim 1, characterized in that During model training, the ten-fold cross-validation method was used to evaluate the model performance in combination with the average root mean square error and average deviation of the prediction results, and the model with the best comprehensive accuracy was selected as the final inversion model.

Images