Ice lake remote sensing monitoring method

Abstract

The invention discloses a remote sensing monitoring method for an ice lake. The method comprises the following steps: 1) downloading and preprocessing remote sensing data; 2) constructing a composite remote sensing index CI based on the NDWI and the DNSI; 3) preprocessing a composite remote sensing index (CI); 4) carrying out DEM data downscaling and pixel matching processing; 5) dynamically calibrating a CI threshold value based on the DEM; and 6) ice water classification and area extraction. In order to solve the problems that a single remote sensing index and a fixed threshold method are low in ice lake area extraction and ice water classification precision and poor in universality under the plateau complex terrain condition, a composite remote sensing index is constructed, a terrain factor is introduced to dynamically adjust a classification threshold, a support vector regression method is utilized to improve the fitting precision of a CI threshold and an elevation relationship, and the classification accuracy is improved. And the precision of extraction of the ice lake area, ice water classification and discrimination and the like is effectively improved. According to the method, the space adaptability and the monitoring precision of a glacial lake classification result are remarkably improved, and technical support is provided for evolution analysis and disaster early warning of the glacial lake in the plateau area.

Description

Technical Field

[0001] This invention relates to remote sensing monitoring technology for glacial lakes, belonging to the field of remote sensing technology. Specifically, it provides a remote sensing monitoring method for glacial lakes that can accurately identify the ice-water boundary and extract the area of glacial lakes in plateau regions. It is suitable for scenarios such as analyzing the dynamic changes of glacial lakes and monitoring disaster risks. Background Art

[0002] Glacial lakes, unique water bodies in the plateau, have boundaries that directly reflect glacier melt, water volume fluctuations, and climate change. Extracting glacial lake boundaries is crucial for assessing glacial lake outburst flood risks, developing disaster prevention and mitigation measures, and implementing water resource management. With the continuous advancement of remote sensing technology, automated methods for extracting glacial lake boundaries based on remote sensing have become a research hotspot. Accurately and efficiently extracting glacial lake boundaries has become a key component of remote sensing monitoring of the plateau.

[0003] Currently, remote sensing index methods, due to their simple structure and efficient computation, have been widely used in glacial lake extraction and water and ice surface classification. Common water identification indices include the Normalized Difference Water Index (NDWI), Modified Normalized Difference Water Index (MNDWI), and Automatic Water Index (AWEI). These indices primarily utilize the differences in water reflectance between visible and near-infrared wavelengths. Common indices for ice and snow identification include the Normalized Difference Snow Index (NDSI), Normalized Difference Ice Index (NDII), and Snow Cover Index (SCI). These primarily reflect ice and snow cover by combining green and shortwave infrared light. In practical applications, different indices exhibit varying sensitivity to their targets. NDWI is sensitive to water but performs poorly in mixed ice and snow areas. NDSI can identify snow surfaces but can be easily confused with clear water. MNDWI and AWEI have advantages in highly reflective backgrounds. Furthermore, some studies have introduced improvements such as spectral angle analysis and principal component analysis (PCA) to enhance the stability of indices in complex backgrounds. However, most of these methods still rely on fixed thresholds and lack linkage adjustment mechanisms with terrain and temporal variation factors.

[0004] Current remote sensing monitoring of glacial lakes mostly relies on a single remote sensing index (such as NDWI and NDSI) combined with a fixed threshold to identify and classify glacial lake boundaries. While this approach has some applicability under specific conditions, it exhibits significant limitations when faced with complex and changing terrain environments and seasonal climate changes. On the one hand, such methods generally lack the ability to respond to factors such as topographic undulations and the temporal variability of water conditions, making them difficult to adapt to the dynamic changes in glacial lake boundaries across different time periods, elevation zones, and even remote sensing imagery. On the other hand, traditional index thresholds often rely on manual experience, lack a scientifically sound dynamic calibration mechanism and regional adaptation strategy, and are easily affected by image lighting conditions, sensor differences, and surface cover. This results in large fluctuations and low accuracy in ice-water classification results, making it difficult to meet the actual needs of rapid early warning and high-precision monitoring of glacial lake hazards. Therefore, there is an urgent need to develop a new glacial lake monitoring method that integrates multi-source remote sensing index information, has dynamic calibration capabilities, and can fully adapt to complex terrain conditions, so as to achieve automation, high precision and strong robustness in the process of glacial lake boundary extraction, and provide more scientific and reliable data support for glacial lake outburst risk warning, water resources management and climate change research. Summary of the Invention

[0005] To address the problems of existing glacial lake remote sensing monitoring methods, such as low ice-water boundary recognition accuracy, poor adaptability of fixed thresholds, and lack of terrain adjustment capabilities, the present invention provides a glacial lake remote sensing monitoring method. The objectives of the present invention are achieved through the following technical solutions:

[0006] A remote sensing monitoring method for glacial lakes is based on optical remote sensing images and digital elevation models (DEMs). This method constructs a composite remote sensing index using NDWI and NDSI. Combined with the downscaled DEM, the method dynamically calibrates the threshold using the SVR method. This method simultaneously achieves accurate extraction of glacial lake area and fine classification of ice and water. The method specifically includes the following steps:

[0007] A glacial lake remote sensing monitoring method comprises the following steps:

[0008] 1) Data download and preprocessing:

[0009] Acquire multi-source optical remote sensing images covering the target glacial lake area at different phases, perform atmospheric and geometric corrections, and obtain the original DEM data of the target glacial lake area.

[0010] The proposed method relies primarily on optical remote sensing data from Landsat, Sentinel-2, and other satellites. First, remote sensing imagery data covering a specific time period must be downloaded from a remote sensing data platform. The downloaded data must include time series images of the glacial lake area. After downloading, the imagery undergoes necessary preprocessing, including atmospheric and geometric corrections, to ensure spatial accuracy and spectral consistency.

[0011] 2) Construction of composite remote sensing index:

[0012] A composite remote sensing index (CI) is constructed by combining the ratio of the NDWI and NDSI remote sensing indices. The NDWI can be used to enhance the identification of water bodies, while the NDSI helps distinguish ice from water. The CI reflects the comprehensive spectral characteristics of the ice-water mixing zone and is used to enhance the identification of glacial lakes.

[0013] 3) Extraction of glacial lake area based on CI index:

[0014] The CI layer was standardized and preprocessed, and the standardized CI image was threshold segmented using the Otsu maximum inter-class variance method to obtain the optimal glacial lake boundary threshold t, and the overall glacial lake area including the ice surface and water body was extracted.

[0015] This step aims to extract the overall extent of the glacial lake region, including the combined coverage of water and ice, based on the Composite Remote Sensing Index (CI) layer. The preprocessed CI layer achieves a uniform numerical scale and significantly improves the index's ability to discriminate between glacial and water areas. Furthermore, the standardized CI layer is segmented using the Otsu maximum inter-class variance method, automatically determining an optimal global threshold to extract the entire glacial lake region (including both water and ice pixels). The resulting extraction forms a glacial lake mask layer, which serves as input for the subsequent ice-water classification step based on dynamic elevation calibration, laying the data foundation for further distinguishing ice from water.

[0016] 4) DEM data downscaling and pixel matching processing:

[0017] The DEM data is resampled using the cubic interpolation method to achieve pixel-level spatial alignment with the CI layer. To achieve pixel-level matching between the 30m resolution DEM data and the 10m resolution CI layer, this paper proposes to downscale the DEM data using the cubic interpolation method, increasing its spatial resolution to 10m, consistent with the CI, to ensure spatial position consistency and the feasibility of data fusion.

[0018] 5) Dynamic determination of ice-water classification thresholds based on DEM:

[0019] Based on the downscaled DEM layer, the natural breakpoint method was used to segment the regional elevation into several elevation intervals. The corresponding CI value set was extracted within each interval. The CI values were classified into ice and water categories using the K-means clustering algorithm, and the optimal CI threshold for ice and water classification was determined for each elevation interval. A sample set between the center elevation of each interval and the CI threshold was constructed, and the elevation-CI threshold mapping function was established using support vector regression (SVR). The SVR model was constructed using the radial basis kernel function (RBF), and the model parameters, including the penalty factor C and the kernel function parameter γ, were automatically determined through K-fold cross-validation.

[0020] The specific operations are as follows:

[0021] First, the downscaled DEM layer is divided into elevation intervals, and the natural breakpoint method (Jenks NaturalBreaks) is used to optimally partition the DEM data to ensure that the elevation difference within the interval is minimized and the elevation difference between intervals is maximized. The optimization objective function of the natural breakpoint method is:

[0022]

[0023] Among them, h i is the elevation value, μ j is the mean of the j-th elevation interval, H j represents the jth interval, and k represents the total number of elevation intervals. The value of k (the total number of elevation intervals) is typically flexibly set based on the terrain complexity and sample distribution of the study area and is an adjustable parameter in the natural breakpoint method. This method ensures that the elevation interval divisions fully reflect topographic variations and optimally zoning the elevation characteristics of different regions, thus providing a clear elevation structure foundation for subsequent ice-water classification.

[0024] Within each elevation interval, a corresponding set of CI values is extracted. These CI values are then classified into ice and water using the K-means clustering algorithm, thereby determining the optimal CI threshold for ice and water classification for each elevation interval. The K-means clustering algorithm performs optimal clustering using the following objective function:

[0025]

[0026] Among them, r ij Is an indicator function, indicating that the sample x i Whether it belongs to the jth category, μ j is the center of the jth class, n represents the total number of samples, that is, the total number of CI layer pixels obtained in step 4), and k represents the total number of elevation intervals after the optimal partitioning of the DEM data; ‖x i -μ j ‖ represents sample x iTo the cluster center μ j The algorithm iteratively optimizes the class label of each data point to minimize the distance from all data points to the center of their class, thereby effectively classifying CI values. The goal of K-means clustering is to classify CI values into two categories: ice and water. The clustering results are used to generate representative ice-water classification thresholds for each elevation interval.

[0027] Finally, support vector regression (SVR) was used to establish an elevation-to-threshold mapping function based on the central elevation of each elevation interval and the corresponding CI threshold sample. SVR is a nonlinear method widely used in regression problems. It can improve the model's ability to fit complex nonlinear relationships by mapping the raw data into a high-dimensional space. By minimizing structural risk, the SVR method can effectively avoid overfitting while maintaining prediction accuracy, thereby providing a precise ice-water classification threshold for each elevation value.

[0028] The objective function of SVR is as follows:

[0029]

[0030] The constraints for the regression are:

[0031] T i -<ω,Φ(h i )>-b≤∈+ξ i ,

[0032] In the above formula: ‖ω 2 ‖ is the square norm of the weight vector, representing the complexity of the regression function; k represents the total number of elevation intervals after the optimal partitioning of the DEM data; i represents the interval number; ξ i and is a slack variable, used to allow a certain prediction error; C is a penalty coefficient, used to control the degree of error penalty; T i Indicates the CI threshold corresponding to each elevation interval; <ω,φ(h i )> is the inner product of the elevation value after mapping by the nonlinear kernel function and the regression function, b is the bias term, ∈ is the regression tolerance, and a certain range of errors is allowed.

[0033] Specifically in the application of the present invention, h i is the central elevation of each elevation interval, indicating the typical representative value of different elevation intervals. When performing SVR modeling, the CI threshold T obtained from step 5) is used. i The corresponding elevation value h i Construct elevation-CI threshold sample sets. These sample sets include elevation and corresponding ice-water classification thresholds, which serve as the input and target output data of SVR.

[0034] SVR uses these samples to train a regression model to fit the relationship between elevation and ice-water classification threshold, thereby forming an elevation-CI threshold mapping function:

[0035] T=f(h)

[0036] Among them, f(h) is the function calculated by the SVR model, which represents the ice-water classification threshold T corresponding to any given elevation value h. In the regression process, the SVR method uses the radial basis function (RBF) kernel for nonlinear projection, and its kernel function form is:

[0037] K(h i ,h j )=exp(-γ‖h i -h j ‖ 2 )

[0038] Among them, γ is the parameter of the RBF kernel function, which controls the width of the kernel function, ‖h i -h j ‖ is the Euclidean distance between two elevation points. Through this nonlinear mapping, SVR can effectively capture the complex relationship between elevation and ice-water classification threshold.

[0039] Finally, the elevation-threshold mapping function obtained by SVR can be used to calculate the dynamic classification threshold according to the elevation value of each pixel. i Combined with CI layers, this method allows for precise ice and water classification. This method significantly improves the accuracy of glacial lake remote sensing classification through dynamic calibration of elevation features, and adapts to changes in different terrain conditions.

[0040] 6) Detailed classification of ice and water surfaces:

[0041] Based on the elevation-CI threshold mapping function, the elevation value of each pixel within the glacial lake area is extracted and substituted into the function to calculate the corresponding CI threshold. This is then compared with the pixel CI value to achieve ice-water classification. The result is a high-resolution, pixel-by-pixel ice-water distribution map, enabling precise remote sensing delineation of the ice surface and water body of the glacial lake.

[0042] Combined with the elevation-CI threshold mapping relationship established in step 5), the ice surface and water surface of the standardized CI layer obtained in step 3) can be finely distinguished. Specifically, for each image pixel to be classified, its corresponding elevation value h is first extracted. i , and input it into the support vector regression model to calculate the dynamic classification threshold T it matches i Then, the dynamic threshold T i and the CI value C of the current pixel iMake a comparison and make a judgment based on the following principles:

[0043]

[0044] Finally, based on the above dynamic classification method, the ice and water classification results of each pixel in the entire remote sensing image were obtained, that is, a high-resolution, pixel-by-pixel distribution map of glacial lakes and ice and water.

[0045] In a further solution, the formula for constructing CI in step 2) is as follows:

[0046]

[0047]

[0048] Green, NIR, and SWIR represent the reflectance of the green, near-infrared, and short-wave infrared bands of the remote sensing data, respectively. Reflectance is a unitless quantity with a value range between 0 and 1. It is extracted from the remote sensing data processing results obtained in step 1).

[0049] In a further approach, in step 3), the standardization preprocessing step involves first screening valid pixels in the CI layer constructed and calculated in the previous step, removing pixels with CI values less than 0 to eliminate interference from non-water bodies and ice surfaces such as cloud shadows, bare ground, and snow. The remaining valid pixels are then normalized, linearly stretching the CI values to the range of 0 to 1 to enhance numerical consistency and comparability across different time series and sensor images.

[0050] In a further solution, in step 3), the formula of Otsu's maximum inter-class variance method is as follows:

[0051]

[0052] Where t is the glacial lake boundary threshold, which is the dividing value between the standardized CI value and the glacial lake area (ice + water) and the non-glacial lake area; The inter-class variance is a measure of inter-class differences, used to assess the effectiveness of distinguishing glacial lakes from non-glacial lakes at the current threshold; a larger value indicates better performance. ω0(t) and ω1(t) represent the proportion of pixels in the standardized CI layer that belong to and do not belong to glacial lakes (ice and water), respectively. μ0(t) and μ1(t) represent the average CI values of pixels that belong to and do not belong to glacial lakes (ice + water), respectively. By maximizing the inter-class variance between glacial lakes and background, this method effectively avoids bias caused by empirical assumptions and improves the objectivity and robustness of glacial lake boundary extraction.

[0053] In a further solution, the cubic interpolation method in step 3) is: construct a cubic fitting function using the surrounding DEM values at the pixel to be interpolated to complete the spatial refinement of the elevation data and ensure spatial consistency with the remote sensing data.

[0054] Specifically, at each DEM target pixel location to be interpolated, select the four adjacent known pixel values and construct a cubic interpolation function for fitting. The interpolation function form is:

[0055] f(x)=a3(x-x0) 3 +a2(x-x0) 2 +a1(x-x0)+a0

[0056] Where x0 is the reference point, and a0 to a3 are coefficients obtained through least squares fitting. In two-dimensional space, this interpolation operation is performed in both the x and y directions to obtain the elevation value of the target pixel. After interpolation, all DEM data are resampled to 10m resolution and spatially aligned with the CI layer to achieve precise pixel-level matching. This step provides critical terrain information for subsequent CI threshold setting and glacial lake area extraction based on dynamic DEM calibration.

[0057] Beneficial effects of the present invention:

[0058] The method of the present invention is based on optical remote sensing data such as Landsat and Sentinel-2. By introducing digital elevation model (DEM) information, the threshold of the composite remote sensing index (CI) constructed based on NDWI and NDSI is dynamically adjusted. It can automatically identify the boundaries of glacial lakes under different terrain conditions and time stages, and improve the accuracy and stability of ice and water classification.

[0059] This invention effectively solves the problems of traditional methods' inability to adaptively adjust thresholds, reliance on empirical judgment for recognition effects, and susceptibility of boundary extraction to environmental interference. It has the advantages of high automation, strong adaptability, and strong scalability, and is particularly suitable for dynamic monitoring of glacial lake areas under complex terrain conditions in plateau regions. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] The present invention will be further described below with reference to the accompanying drawings and examples.

[0061] Figure 1 This is a technical flow chart of a glacial lake remote sensing monitoring method of the present invention.

[0062] Figure 2 This is the result of glacial lake boundary extraction and ice water classification in Example 1. DETAILED DESCRIPTION

[0063] Example 1:

[0064] A remote sensing monitoring method for glacial lakes is described. In this example, the monitoring area is Selin Co Glacial Lake, a large glacial lake located on the Qinghai-Tibet Plateau and a typical plateau lake. The ecological environment surrounding the lake is fragile and sensitive, significantly affected by climate change, glacial retreat, and water level fluctuations.

[0065] Remote sensing data introduction:

[0066] In embodiment 1, the optical remote sensing image data used is Sentinel-2, and the data source is the public remote sensing image released by the European Space Agency (ESA), which can be obtained and processed for free, and the imaging dates are November 15, 2023, December 21, 2023, January 14, 2024 and January 21, 2024 respectively. The Sentinel-2 satellite is equipped with a multispectral imager (MSI), which can obtain remote sensing data of 13 bands, covering a wide range of bands from visible light, near infrared to shortwave infrared, with a spatial resolution of 10 meters, which can accurately capture the lake boundary and its changes, especially during winter and spring, and can provide the fine change information of the glacial lake. The specific data source is the public remote sensing image released by the European Space Agency (ESA), which can be obtained and processed for free. This makes it possible for researchers to use these image data to carry out multi-temporal analysis in the monitoring of Selin Co Glacial Lake, and to track the key information such as ice water changes, area changes, etc. of the lake in real time.

[0067] Digital elevation model (DEM) is a model that digitally represents the ground height and is widely used in geographic information systems (GIS). In embodiment 1, DEM data is derived from ASTER GDEM (Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Digital Elevation Model). ASTER GDEM is a global elevation data set released by NASA and the University of Tokyo, Japan, and is generated using remote sensing image data from ASTER satellites. The spatial resolution of this data set is 30 meters, covering the entire globe, including polar regions, and data accuracy is relatively efficient.

[0068] The following are the specific steps of Example 1:

[0069] 1) Data acquisition and preprocessing:

[0070] Acquire multi-source optical remote sensing images covering the target glacial lake area at different phases, perform atmospheric and geometric corrections, and obtain DEM data for the target glacial lake area.

[0071] First, we downloaded four Sentinel-2 Level-1C images covering Selin Co Glacial Lake from the Copernicus Open Access Hub platform. The imaging dates are: November 15, 2023, December 21, 2023, January 14, 2024, and January 21, 2024.

[0072] After downloading the Sentinel-2 Level-1C imagery, we first perform atmospheric correction using the Sen2Cor tool, converting it into a surface reflectance product (Level-2A). Based on a physical radiation transfer model, Sen2Cor effectively removes interference from atmospheric aerosols and water vapor, improving the accuracy of spectral metrics like the CI index.

[0073] Subsequently, each of the four images underwent geometric correction to minimize spatial deviations between them. To ensure spatial consistency between the multi-temporal images, a collective control point (GCP) registration method based on Google Earth high-resolution imagery was used. Each image underwent secondary geometric correction and sub-pixel registration to ensure perfect pixel-level alignment, thereby improving the spatial accuracy of subsequent glacial lake boundary extraction and change detection.

[0074] 2) Construction of Composite Remote Sensing Index (CI):

[0075] The Normalized Difference Water Index (NDWI) and the Normalized Difference Snow Index (NDSI) are combined by their ratio to enhance the water response while reducing the confusion between snow and ice. In this step, the CI is constructed by first extracting Bands 3, 8, and 11 from four preprocessed Sentinel-2 images. NDWI and NDSI values are calculated for each image. CI values are then calculated at the pixel level. To further enhance the CI's sensitivity to water-ice transition characteristics in typical glacial lake regions, the Selin Co Lake area is divided into five typical subregions during processing: the southwest shore mudflats, the northeastern bay, the northern part of the main lake area, the glacial lake inlet, and the nearshore transition zone. 500 to 800 representative pixels are selected from each subregion for subsequent extraction of the CI's typical range and verification of the ice-water response.

[0076] 3) Extraction of glacial lake area based on CI index:

[0077] The CI layer was standardized and preprocessed, and the standardized CI image was threshold segmented using the Otsu maximum inter-class variance method to obtain the optimal glacial lake boundary threshold t, and the entire glacial lake area including the ice surface and water body was extracted;

[0078] Based on the CI layer generated in step 2), the overall coverage of the glacial lake in the Selin Co area is extracted, including the comprehensive area of water bodies and ice surfaces. First, the CI layer is screened for valid pixels, pixels with CI values less than 0 are eliminated, and invalid information interference caused by high reflective interference such as cloud shadows, bare ground, and snow is eliminated. Subsequently, the CI values of all valid pixels are linearly normalized so that they fall uniformly into the numerical range of 0 to 1. The CI layer is then binary segmented using the Otsu maximum inter-class variance method to automatically determine the optimal threshold t of the glacial lake boundary. The resulting binary mask layer identifies all pixels with CI greater than the threshold t as glacial lake areas, forming a glacial lake extraction mask map covering the entire Selin Co area.

[0079] 4) DEM data downscaling and pixel matching processing:

[0080] The original 30-meter DEM was downscaled using the cubic interpolation method. For each 10-meter resolution target pixel to be generated, the elevation value at the target location was obtained by constructing a cubic interpolation function in the x and y directions based on the 16 surrounding 30-meter pixels.

[0081] To ensure the accuracy of spatial registration of the resampled data, the interpolated DEM was ensemble-rectified using the projection information of the Sentinel-2 imagery as a standard benchmark, ensuring full consistency with the CI layer in both projected coordinate system and spatial distribution. The resulting 10m-resolution DEM was aligned with the CI layer at the pixel level, with spatial deviations within a single pixel, providing a foundation for high-precision fusion.

[0082] 5) Dynamic determination of ice-water classification thresholds based on DEM

[0083] First, the elevation data for the Selin Co glacial lake area was automatically partitioned using the Jenks Natural Breaks method. Based on prior knowledge of Selin Co Lake, the elevation intervals were divided into 12 levels to maximize topographic homogeneity within the intervals and enhance the representation of elevation differences between intervals. Subsequently, the corresponding pixel set in the CI layer was extracted for each elevation interval. To improve classification accuracy, at least 200 valid pixels were selected from each interval. After considering regional distribution and spectral stability, the CI values were classified into two categories using the K-means clustering algorithm, representing ice surface and water surface pixels, respectively. Based on this, the optimal CI threshold for ice-water classification was determined for each elevation interval.

[0084] A support vector regression (SVR) model was used to construct a mapping function between the ice-water classification threshold and elevation. The central elevation value of the elevation interval and the corresponding CI threshold were used as training samples to input the model. The radial basis function (RBF) kernel function was used for nonlinear fitting to capture the complex relationship between the changes in ice-water reflectivity under different elevation conditions. During the model training process, the kernel width σ was set to 0.3, γ was 5.56, and the penalty coefficient C was set to 100. The model performed stably in the five-fold cross validation, and the fitting accuracy R 2 It is 0.92, indicating that there is a good mapping relationship between elevation and threshold.

[0085] 6) Detailed classification of ice and water surfaces

[0086] Based on the elevation-CI threshold mapping relationship established in step 5), a fine classification of ice surface and water surface of each pixel in the glacial lake image is achieved. The results are shown in the attached figure. Figure 2 Figures (a)-(d) represent the extraction results of Selin Co Lake on November 15, 2023, December 21, 2023, January 14, 2024, and January 21, 2024, respectively. It can be seen that the outline of the ice lake on different dates is relatively clear. The extraction results of different periods are analyzed: Figure (a) shows the ice and water classification extraction during the non-freezing period. At this time, the lake surface has not yet frozen, and the water extraction results are accurate; Figure (b) shows the ice and water classification of the lake surface just entering the freezing period. It can be seen that the lake has begun to freeze. The relatively low-lying areas such as the northern, southern, and eastern coasts of the lake area first freeze. The ice-water boundary is clear and the classification is accurate. Figure (c) shows the ice and water classification of the lake surface during a certain period of time during the freezing period. At this time, only the higher-lying areas such as the western coast of the lake area still have some water surface, while the rest of the lake area is frozen. The ice-water boundary is clear and the classification is accurate; Figure (d) shows the lake surface entering the completely frozen stage. From the above analysis, it can be seen that the overall monitoring situation of the ice lake monitoring work in Example 1 of the present invention is relatively accurate, reflecting the changes in ice and water on the surface of Selin Co Lake during the freezing period.

[0087] Finally, it should be noted that the above is only used to illustrate the technical solution of the present invention and is not limiting. Although the present invention is described in detail with reference to the preferred arrangement scheme, ordinary technicians in this field should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A glacial lake remote sensing monitoring method, characterized by: The following steps are involved: 1) Data acquisition and preprocessing: Acquire multi-source optical remote sensing images covering the target glacial lake area at different phases and perform atmospheric and geometric corrections; and obtain DEM data for the target glacial lake area; 2) Construction of composite remote sensing index: The composite remote sensing index CI is constructed based on the ratio of the normalized water index NDWI to the normalized snow difference index NDSI; 3) Glacial lake area extraction based on CI index: The CI layer is standardized and preprocessed, and the standardized CI image is threshold segmented using the Otsu maximum inter-class variance method to obtain the optimal glacial lake boundary threshold and extract the entire glacial lake area including the ice surface and water body; 4) DEM data downscaling and pixel matching processing: The DEM data is resampled and downscaled using the cubic interpolation method to achieve pixel-level spatial alignment with the CI layer; 5) Dynamic calibration of ice-water classification thresholds based on DEM: Based on the downscaled DEM layer, the natural breakpoint method is used to segment the regional elevation into several elevation intervals. The corresponding CI value set is extracted within each interval. The CI values are classified into ice and water categories using the K-means clustering algorithm. The optimal CI threshold for ice-water classification is determined for each elevation interval. A sample set between the center elevation of each interval and the CI threshold is constructed, and the support vector regression method (SVR) is used to establish the elevation-CI threshold mapping function. 6) Fine classification of ice and water surfaces: Based on the elevation-CI threshold mapping function, for each pixel in the glacial lake area, its elevation value is extracted and substituted into the function to calculate the corresponding CI threshold, which is then compared with the pixel CI value to achieve ice and water classification.

2. The glacial lake remote sensing monitoring method according to claim 1, characterized in that: The CI value formula in step 2) is as follows: Where Green, NIR, and SWIR represent the reflectance of the green band, near infrared band, and shortwave infrared band of remote sensing data, respectively.

3. The glacial lake remote sensing monitoring method according to claim 1, characterized in that: In step 3), the step of performing standardization preprocessing on CI includes: removing pixels with CI values less than 0, retaining valid data, and performing normalization processing to stretch the CI value to the range of 0 to 1.

4. The glacial lake remote sensing monitoring method according to claim 1, characterized in that: In step 4), the 30m resolution DEM data is specifically matched with the 10m resolution CI layer at the pixel level.

5. The glacial lake remote sensing monitoring method according to claim 1, characterized in that: In step 4), the cubic interpolation method is to construct a cubic fitting function at the pixel to be interpolated using the surrounding DEM values to complete the spatial refinement of the elevation data and maintain the spatial consistency of the remote sensing data.

6. The glacial lake remote sensing monitoring method according to claim 1, characterized in that: In step 5), the SVR model is constructed using a radial basis kernel function (RBF), and the model parameters include a penalty factor C and a kernel function parameter γ, which are automatically determined through K-fold cross validation.

7. The glacial lake remote sensing monitoring method according to claim 1, characterized in that: In step 6), for each image pixel to be classified, extract its corresponding elevation value h i , and based on the elevation-CI threshold mapping function, calculate the dynamic classification CI threshold T that matches it i ; Set the threshold T i and the CI value C of the current pixel i Make a comparison and make a judgment based on the following principles: Based on the above dynamic classification, the ice and water classification results of each pixel in the entire remote sensing image are obtained, and the ice and water distribution map of the glacial lake is obtained pixel by pixel.

Images