Method for monitoring form information of ice lake in area lacking data based on satellite remote sensing

Abstract

The invention discloses a satellite remote sensing-based ice lake form information monitoring method for a region lacking data. The method comprises the following steps of: 1) acquiring and preprocessing height measurement satellite data; 2) extracting the water level of the glacial lake based on a self-adaptive threshold waveform redetermination method; 3) acquiring and preprocessing multi-source optical satellite data; 4) calculating a water body index; 5) extracting an ice lake contour based on a machine learning method; and 6) constructing a water level-area-water volume relationship. According to the method, the precision of extracting the water level by height measurement satellite data is improved by combining an adaptive threshold waveform redetermination method, the precision of extracting the contour of the ice lake is improved by combining several traditional water body indexes and by means of a random forest classification method, the blank of ice lake form information monitoring work in areas lacking data is made up to a certain extent, and the method is suitable for popularization and application. And efficient technical support is provided for evolution analysis and disaster early warning of the ice lake in the plateau area.

Description

Technical Field

[0001] The present invention belongs to the field of remote sensing technology and geographic information technology, and relates to remote sensing monitoring technology for glacial lakes in data-deficient areas, and specifically to a method for monitoring glacial lake morphological information in data-deficient areas based on satellite remote sensing. Background Art

[0002] Glacial lakes are water bodies formed in alpine regions by the confluence of glacial meltwater, ice obstruction, or moraine dams. Widely distributed across the Qinghai-Tibet Plateau, the Himalayas, and other polar and high-altitude regions, they represent a vital freshwater resource. In recent years, with global warming and glacial retreat accelerating, the number and area of glacial lakes have continued to grow, leading to an increasingly severe glacial lake outburst flood (GLOF) problem, threatening downstream ecological and infrastructure security. Glacial lake morphology is a key component of monitoring and risk assessment. This information, encompassing the contours, water levels, and volume of glacial lakes, reflects their spatial extent, elevation, and water storage, respectively. It is crucial for GLOF early warning, regional water resource assessment, and climate change research.

[0003] Due to complex terrain, high altitude, and harsh climate conditions, glacial lakes are often located in remote and inaccessible areas, making traditional ground-based monitoring difficult to implement in a stable and long-term manner. Equipment such as water level gauges and flow meters are difficult to deploy, have high maintenance costs, and have limited coverage. As a result, many glacial lakes remain in a "no data" state for long periods of time, with a serious lack of morphological information. In particular, in the prevention and control of sudden and destructive GLOF risks, the lack of continuous and uniform monitoring data has become a key bottleneck restricting the development of disaster assessment and early warning systems. In recent years, remote sensing technology has become an important supplement to glacial lake monitoring due to its advantages of large-scale, periodic, and non-contact observation.

[0004] Remote sensing monitoring of glacial lake morphology primarily focuses on boundary extraction, water level inversion, and water volume estimation. However, existing methods have significant shortcomings. For boundary extraction, current methods rely primarily on medium- to high-resolution optical imagery, using threshold segmentation and spectral feature analysis to identify water body edges. However, due to interference from ice, snow, and complex terrain, this can easily lead to misidentification or omission. Especially in high-albedo environments, it is often difficult to accurately determine the true lake boundary. For water level inversion, mainstream methods rely on digital elevation models (DEMs) to infer water surface elevation. However, the limited resolution and accuracy of existing DEM data make water level extraction susceptible to terrain fluctuations and noise, resulting in large errors in water level values. Furthermore, while some studies have incorporated satellite altimetry data, such as the Jason-1 / 2 / 3 series and the SWOT satellites, which use radar altimeters to obtain surface reflection signals, these data still face challenges during processing, such as waveform interference and offset of preset tracking gates. Traditional re-calibration methods have failed to effectively overcome these challenges, compromising the final water level accuracy and failing to meet the precise requirements for glacial lake disaster warning and water resource management. In terms of water volume estimation, most methods use a simplified water depth model and simply multiply the lake area by the water depth to estimate the water volume. However, this method fails to fully consider the complex geometric shape of glacial lakes and changes in lake bottom topography, and often ignores factors such as edge slope and local depressions, thereby limiting the reliability and accuracy of the estimation results.

[0005] Overall, most studies focus solely on a single factor, lacking a coordinated monitoring approach for these factors. This has hindered comprehensive, spatiotemporal monitoring of glacial lakes. Existing monitoring methods are significantly limited in remote sensing monitoring of glacial lakes in high-altitude, cold, and data-deficient regions. A comprehensive solution is urgently needed that leverages multi-source data, combines machine learning, and advanced signal processing techniques, to improve the accuracy and practicality of extracting glacial lake morphological information. Summary of the Invention

[0006] To address the aforementioned technical deficiencies, the present invention provides a method for monitoring glacial lake morphology in data-deficient areas based on satellite remote sensing. This method, through a multi-factor collaborative monitoring approach, enables systematic, spatiotemporal monitoring of glacial lakes, thereby improving the accuracy and practicality of glacial lake morphology information extraction. The present invention achieves this objective through the following technical solutions:

[0007] A method for monitoring glacial lake morphology in data-deficient areas based on satellite remote sensing includes the following steps:

[0008] 1) Altimetry satellite data acquisition and preprocessing: Altimetry satellite data preprocessing includes denoising and spatiotemporal registration.

[0009] 2) Using the adaptive threshold waveform re-gate method to extract glacial lake water levels from ATR-preprocessed altimetry satellite data: an adaptive differential sequence is used to identify the leading edge of the waveform and determine the threshold for detecting the leading edge. The range of the sub-waveform is further expanded by weighted re-gates. An adaptive threshold is calculated to determine whether a valid re-gate position exists in the waveform. A distance correction value is obtained based on the time of the re-gate position to modify the water level measurement formula and obtain the final water level.

[0010] 3) Multi-source optical satellite data acquisition and preprocessing: The preprocessing of multi-source optical satellite data includes atmospheric correction and geometric correction;

[0011] 4) Calculation of water index based on pre-processed multi-source optical satellite data The water index includes Normalized Difference Water Index (NDWI), Modified Normalized Difference Water Index (MNDWI), Automatic Water Extraction Index (AWEI), Shadow Formula (AWEI) and other water indexes. sh ) and the shadowless formula (AWEIn sh ) two types;

[0012] 5) Extraction of glacial lake contours based on machine learning methods: Random forest classification method is used to extract the contours of glacial lakes using four water body indices: NDWI, MNDWI, and AWEI. sh 、AWEI nsh The information is used as input to accurately identify the outline of the glacial lake;

[0013] 6) Establishing the water level-area-water volume relationship: Obtain water level data from step 2), obtain contour changes from step 5), and further calculate area changes. First, establish the water level-area relationship. Assuming the lake is conical or truncated, establish the area-water volume relationship, and finally obtain the water level-area-water volume relationship.

[0014] In a further solution, the altimetry satellite data used in step 1) is from the Jason series of satellites, jointly developed by the National Aeronautics and Space Administration (NASA) and the French National Center for Space Studies (CNES). These satellites are primarily used to monitor global ocean surface height. The water surface height data provided by these satellites, available through NASA's Ocean Atmosphere Data System (PODAAC), covers the glacial lake region and provides crucial data support for glacial lake water level monitoring.

[0015] After acquiring Jason-1 / 2 / 3 altimeter data, the raw data undergoes two preprocessing steps: denoising and spatiotemporal registration. Gaussian filtering is used to remove noise from the acquired data, eliminating anomalous signals caused by snow, ice, and other environmental factors. To ensure spatiotemporal consistency, the altimeter data must be registered with other remote sensing imagery (such as high-resolution optical imagery) to ensure consistency in spatial coordinates and temporal references. This step utilizes high-resolution remote sensing imagery from Google Earth for registration.

[0016] In a further solution, in step 2), in the extraction of glacial lake water levels, traditional waveform retracking methods may be affected by environmental factors such as high noise, unstable ice surface reflection, and waves. This step introduces the adaptive threshold waveform retracking method (Adaptive Threshold Retracking, ATR). The core idea of this method is to accurately identify the waveform retracking gate position by introducing adaptive thresholds and time windows, thereby optimizing the accuracy of water level extraction.

[0017] First, extract the sub-waveform from the echo power sequence and apply the adaptive difference sequence (ADS) to more accurately identify the leading edge of the waveform. The adaptive difference sequence (ADS) calculation formula is:

[0018]

[0019] Where: P(i) and P(j) are the echo intensity values of the i-th and j-th sampling points of the echo power sequence respectively; D ADS (i) represents the local difference of the i-th point; k is the local window size used to calculate the local difference; and j represents the index of all sampling points in the range from ik to i + k. By introducing the adaptive difference sequence, we can accurately capture rapidly changing parts of the waveform and identify the waveform leading edge. When the local difference exceeds a certain threshold, it is considered a potential sub-waveform leading edge.

[0020] After extracting the sub-waveform, the threshold is determined. This threshold is used to detect the leading edge of the waveform. The calculation formula is as follows:

[0021] T ADS (i)=α·σ(i)

[0022] Among them: α is a constant used to control the sensitivity of the threshold; σ(i) is the standard deviation of the window where the i-th sampling point is located, reflecting the range of variation of the waveform. ADS (i) Greater than the threshold T ADS(i) is considered as a potential waveform front edge. This method can identify the waveform change point in an environment with weak signals or strong noise, thereby extracting the possible sub-waveform range.

[0023] Next, the range of the sub-waveform is further expanded by weighting the gate. The weighted power difference is calculated as follows:

[0024]

[0025] Where P(i) is the echo power at the current sampling point, max(P[i,j]) and min(P[i,j]) are the minimum and maximum echo powers within the sub-waveform range, respectively. This formula normalizes the power differences between waveforms, focusing on the leading edge signal and improving the ability to identify low-signal waveforms.

[0026] Then, the adaptive threshold is calculated to determine whether there is a valid reset gate position in the waveform. The determination of the reset gate position depends on the fluctuation of the echo signal and the environmental noise. The formula is as follows:

[0027] T r =β·σ window (i)+γ·Noise

[0028] Where: T r is the adaptive threshold; σ window (i) is the standard deviation of the current window, indicating the fluctuation of the echo signal in the area; β and γ are adjustment parameters; Noise is the noise level, indicating the influence of environmental noise. When the signal exceeds the threshold, the point is determined to be the re-gate position.

[0029] The final distance correction formula corrects the water level measurement error and obtains the final water level. The formula is as follows:

[0030] ΔR=2c·(t GR -t GN )

[0031] H=R-ΔR

[0032] Where: c is the speed of light, t GR is the time to reset the door position, t GN is the preset tracking gate time, ΔR is the distance correction value, ΔR is the initially measured distance, and H is the final water level.

[0033] In a further solution, the multi-source optical satellite data used in step 3) includes Sentinel-2 and Landsat series data. Sentinel-2 data is provided by the European Space Agency (ESA) and is available through the Copernicus open access platform, with a spatial resolution of 10 meters. Landsat series data is provided by the United States Geological Survey (USGS) and is available for download through the EarthExplorer platform, with a spatial resolution of 30 meters.

[0034] To ensure image quality, the acquired remote sensing data underwent standardized preprocessing. Atmospheric correction employed the Sen2Cor (for Sentinel-2) and LEDAPS (for Landsat) methods, both based on radiative transfer models, to eliminate interference from atmospheric aerosols, water vapor, and other factors on surface reflectivity. Geometric correction employed precise registration using Google Earth high-resolution remote sensing imagery as a reference, ensuring consistency in spatial positioning across multiple data sources.

[0035] In a further solution, the calculation method in step 4) is as shown in the following formula:

[0036]

[0037]

[0038] AWEI sh =4(Green-SWIR1) / (0.25NIR+2.75SWIR2)

[0039] AWEI nsh =Blue+2.5Green-1.5(NIR+SWIR1)-0.25SWIR2

[0040] Among them: Green represents the green light band of multispectral imaging (520–600 nm), NIR represents the near-infrared band (760–900 nm), and SWIR represents the shortwave infrared band (1550–1750 nm), which includes two bands, SWIR1 and SWIR2, with wavelength ranges of approximately 1550–1750 nm and 2080–2350 nm, respectively.

[0041] Furthermore, the random forest algorithm in step 5) is an ensemble learning method that constructs multiple decision trees for classification and regression. Each tree is trained by randomly selecting a subset of data and features, and the classification result is ultimately determined through a voting mechanism. This method effectively handles high-dimensional data, reduces overfitting, and exhibits strong robustness.

[0042] In the extraction of glacial lake contours, random forest can integrate four water body indices (NDWI, MNDWI, AWEI sh 、AWEI nsh ) information, accurately identifying glacial lake boundaries, adapting to diverse environmental changes, and providing highly accurate classification results. The prediction of each decision tree can be expressed as a function based on four water index features, as shown below.

[0043]

[0044] Where: X j =[NDWI, MNDWI, AWEI sh , AWEI nsh ] are the four water index features input, R i is the i-th leaf node area, I(X j ∈R i ) is the indicator function, which represents the eigenvector X j Whether it belongs to the i-th leaf node, y i is the class label of the leaf node. This formula shows that each tree is based on four water body indices (NDWI, MNDWI, AWEI sh 、AWEI nsh ) to divide the feature space of the sample and classify the sample into the corresponding leaf node area to perform classification prediction. The output of each tree is based on the input feature X j The classification results.

[0045] Finally, the overall classification result of the random forest is determined by voting among multiple decision trees:

[0046] y RF =model(f1(X),f2(X),...,f N (X))

[0047] Where: y RF is the classification result; model(...) is the trained random forest classification model; f i (X) is the classification result of category i. The classification of the ice lake outline is finally determined by voting among all trees.

[0048] In a further solution, in step 6), the water level data is obtained from step 2), and the contour changes are obtained from step 5). Contour changes usually represent changes in the water body area. Therefore, a water level-area relationship can be established first. First, it is necessary to select an appropriate time period and ensure that the time difference between the optical image and the altimetry satellite water level data is as small as possible (usually controlled within 10 days). The extracted local contour changes are reflected in the water body area and paired with the altimetry satellite water level data for the corresponding time period. Since altimetry satellite data provides water surface height and lake water level changes are usually uniform, contour changes are directly related to water level changes. Through regression analysis, a mathematical relationship between area change and water level is established. Generally, this relationship is linear, that is, as the water level rises, the water body area will change linearly. Through the linear regression model, the regression equation that best fits the data can be found, thereby converting area changes into corresponding water level values.

[0049] When constructing area-volume relationships, the lake is typically assumed to be conical or truncated cone-shaped. This means that the relationship between water level and the lake surface area reflected by its contour is typically a quadratic function. Through regression analysis, it is possible to derive a mathematical relationship between water level and area from measured lake area and water level data. Furthermore, this relationship can be used to calculate an integral to determine the relationship between water level and lake volume.

[0050] Beneficial effects of the present invention:

[0051] The present invention takes the morphological information of glacial lakes, such as their contours, water levels, and water storage capacity, as the research object. Based on data from multi-source remote sensing satellites, altimetry satellites, and other sources, combined with machine learning and waveform recalibration methods, a satellite remote sensing-based glacial lake morphological information monitoring method for areas with insufficient data is provided. This method realizes remote sensing monitoring of the morphological information of glacial lakes in areas with insufficient data, and is suitable for scenarios such as dynamic monitoring of glacial lakes and risk analysis.

[0052] The present invention realizes multi-temporal and spatial systematic monitoring of glacial lakes through a multi-factor collaborative monitoring method, thereby improving the accuracy and practicality of glacial lake morphological information extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] The present invention will be further described below with reference to the accompanying drawings and examples.

[0054] Figure 1 This is a technical flow chart of Example 1, a method for monitoring glacial lake morphology information in data-deficient areas based on satellite remote sensing.

[0055] Figure 2 This is a schematic diagram of the geographical location of the Selin Co River Basin in Example 1.

[0056] Figure 3 This is the water level extraction accuracy verification result of the Selin Co Lake in the Selin Co Basin in Example 1.

[0057] Figure 4 This is the calculation result of the water index of the typical lake in the Selin Co Basin in Example 1.

[0058] Figure 5 This is the contour extraction result of the typical lake in the Selin Co Basin in Example 1.

[0059] Figure 6 This is the water level-area-water volume relationship of the typical lake in the Selin Co Basin in Example 1. DETAILED DESCRIPTION

[0060] Example 1:

[0061] A method for monitoring glacial lake morphology in data-deficient areas based on satellite remote sensing. In this embodiment, the monitoring area is the Selin Co River Basin. The following are the specific steps of Example 1:

[0062] 1) Altimetry satellite data acquisition and preprocessing: Altimetry satellite data preprocessing includes denoising and spatiotemporal registration.

[0063] In this embodiment, after obtaining Jason-1 / 2 / 3 altimetry satellite data for the Selinco River Basin from 2010 to 2024 through NASA's Ocean Atmosphere Data System (PO.DAAC), the raw data are preprocessed in two steps: denoising and spatiotemporal registration.

[0064] After acquiring Jason-1 / 2 / 3 altimeter data, the raw data undergoes two preprocessing steps: denoising and spatiotemporal registration. Gaussian filtering is used to remove noise from the acquired data, eliminating anomalous signals caused by snow, ice, and other environmental factors. To ensure spatiotemporal consistency, the altimeter data must be registered with other remote sensing imagery (such as high-resolution optical imagery) to ensure consistency in spatial coordinates and temporal references. This step utilizes high-resolution remote sensing imagery from Google Earth for registration.

[0065] 2) Using the adaptive threshold waveform re-gate method to extract glacial lake water levels from ATR-preprocessed altimetry satellite data: an adaptive differential sequence is used to identify the leading edge of the waveform and determine the threshold for detecting the leading edge. The range of the sub-waveform is further expanded by weighted re-gates. An adaptive threshold is calculated to determine whether a valid re-gate position exists in the waveform. A distance correction value is obtained based on the time of the re-gate position to modify the water level measurement formula and obtain the final water level.

[0066] In this embodiment, water level data is extracted from the results of step 1) based on the adaptive threshold waveform resetting method (ATR). Taking the water level data of Selin Co observed by the Jason-3 altimeter satellite as an example, the actual water level of Selin Co in 2021, 2022 and 2023 is verified. After removing the relative error between the altimeter satellite and the measured water level, the water level accuracy of the Jason-3 altimeter satellite can reach 7 cm (RMSE), R 2 Reached 0.91.

[0067] 3) Multi-source optical satellite data acquisition and preprocessing: The preprocessing of multi-source optical satellite data includes atmospheric correction and geometric correction.

[0068] In the embodiment, the high-resolution optical satellite data used include Sentinel-2 and Landsat series data, etc. Sentinel-2 data is obtained through the Copernicus open access platform, and Landsat series data is obtained through the EarthExplorer platform.

[0069] After downloading, Sentinel-2 Level-1C images are first atmospherically corrected using the Sen2Cor tool and converted 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 such as the CI index. Landsat images are converted into surface reflectance products (Level-2) using the LaSRC atmospheric correction method, effectively removing atmospheric interference and improving the accuracy of spectral indices. Both optical images are geometrically corrected using Google Earth high-resolution remote sensing images as a reference.

[0070] 4) Water index calculation: Calculate four water indexes, including Normalized Difference Water Index (NDWI), Modified Normalized Difference Water Index (MNDWI), Automatic Water Extraction Index (AWEI), Shadow Formula (AWEI sh ) and the shadowless formula (AWEI nsh ) There are two types. The calculation formula is as follows:

[0071]

[0072] AWEI sh =4(Green-SWIR1) / (0.25NIR+2.75SWIR2)

[0073] AWEI nsh=Blue+2.5Green-1.5(NIR+SWIR1)-0.25SWIR2

[0074] Among them: Green represents the green band of multispectral imagery (520–600 nm), corresponding to the B3 band of Sentinel-2 and the B3 band of Landsat series satellites (taking Landsat 8 as an example); NIR represents the near-infrared band (760–900 nm), corresponding to the B8 band of Sentinel-2 and the B5 band of Landsat series satellites (taking Landsat 8 as an example); SWIR represents the shortwave infrared band (1550–1750 nm), including two bands SWIR1 and SWIR2, which correspond to the B11 and B12 bands of Sentinel-2 and the B6 and B7 bands of Landsat series satellites (taking Landsat 8 as an example). The extraction results are shown in the attached figure. Figure 4 .

[0075] 5) Extraction of glacial lake contours based on machine learning methods: Using the random forest classification method, four water body indices (NDWI, MNDWI, AWEI sh 、AWEI nsh ) information as input to accurately identify the outline of the glacial lake.

[0076] Using four water body indices as input, we evenly distributed more than 400 sampling points in the Selin Co watershed and collected 140,000 training samples from Landsat 8 / 9 and Sentinel-2 images from 2021 and 2022. The size of the random forest decision tree is 50, and the input information is AWEI. sh , AWEI nsh , MNDWI and NDWI four water body indices, the goal is to classify each pixel into two categories: water body and non-water body, where water body pixels also include frozen lake surface. The classification accuracy of the trained model (test set) can reach 0.97, and the Kappa coefficient can reach 0.93, which can meet the needs of extracting the lake surface range and shoreline variation range. Typical glacial lake contour extraction is shown in the attached Figure 5 .

[0077] 6) Construction of water level-area-water volume relationship

[0078] The relationship between water level, area and water volume of the 11 major lakes in the Selin Co Basin is as follows: Figure 6As shown in the figure, the 11 major lakes in the Selin Co basin form two main inflow systems. The most important are the lakes connected by the Zhagen Zangbu system, namely, Zhazang Co, Yueqia Co, Mudiladayu Co, Geren Co, Jiaxia Zangbu, Zigui Co, Sirong Zangbu, Wuru Co, Qiagui Co, Zhagen Zangbu, and Selin Co. The second most important are the lakes connected to Selin Co via the Ali Zangbu system, namely, Mujiu Co, Yongzhu Zangbu, Co'e, Ali Zangbu, and Selin Co. This example focuses on extracting lake water regime elements for Geren Co, Zigui Co, Wuru Co, Qiagui Co, Mujiu Co, and Co'e.

[0079] 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 method for monitoring glacial lake morphology in data-deficient areas based on satellite remote sensing, characterized by: The following steps are involved: 1) Altimetry satellite data acquisition and preprocessing: Altimetry satellite data preprocessing includes denoising and spatiotemporal registration; 2) Using the adaptive threshold waveform re-gate method to extract glacial lake water levels from ATR-preprocessed altimetry satellite data: an adaptive differential sequence is used to identify the leading edge of the waveform and determine the threshold for detecting the leading edge. The range of the sub-waveform is further expanded by weighted re-gates. An adaptive threshold is calculated to determine whether a valid re-gate position exists in the waveform. A distance correction value is obtained based on the time of the re-gate position to modify the water level measurement formula and obtain the final water level. 3) Multi-source optical satellite data acquisition and preprocessing: The preprocessing of multi-source optical satellite data includes atmospheric correction and geometric correction; 4) Calculate water index based on pre-processed multi-source optical satellite data: the calculated water index includes normalized difference water index NDWI, improved normalized difference water index MNDWI and automatic water extraction index AWEI, the AWEI is divided into shadow formula AWEI sh and shadowless formula AWEI nsh Two types; 5) Extraction of glacial lake contours based on machine learning methods: Random forest classification method is used to extract the contours of glacial lakes using four water body indices: NDWI, MNDWI, and AWEI. sh 、AWEI nsh The information is used as input to accurately identify the outline of the glacial lake; 6) Construction of water level-area-water volume relationship: Obtain water level data from step 2) and contour changes from step 5). First, establish the water level-area relationship. Assuming that the lake is conical or truncated cone, construct the area-water volume relationship, and finally obtain the water level-area-water volume relationship.

2. The method for monitoring glacial lake morphology in data-deficient areas based on satellite remote sensing according to claim 1, characterized in that: In step 1), Gaussian filtering is used to denoise the acquired altimetry satellite data; and the satellite altimetry data is spatially and temporally aligned with the high-resolution remote sensing imagery from Google Earth.

3. The method for monitoring glacial lake morphology in data-deficient areas based on satellite remote sensing according to claim 1, characterized in that: In step 2), the calculation formula of the adaptive differential sequence ADS is as follows: Where: P(i) and P(j) are the echo intensity values of the i-th and j-th sampling points of the echo power sequence respectively; D ADS (i) is the adaptive difference sequence, which is used to represent the local difference of the i-th point; k is the local window size, which is used to calculate the local difference, and j represents the index of all sampling points in the range from ik to i+k.

4. The method for monitoring glacial lake morphology in data-deficient areas based on satellite remote sensing according to claim 1, characterized in that: In step 2), the range of the sub-waveform is further expanded by weighting the gate, and the calculation formula of the weighted power difference used is as follows: Where: P(i) is the echo power of the current sampling point, max(P[i,j]) and min(P[i,j]) are the minimum and maximum echo powers within the sub-waveform range, respectively.

5. The method for monitoring glacial lake morphology in data-deficient areas based on satellite remote sensing according to claim 1, characterized in that: In step 2), the adaptive threshold is calculated to determine whether there is a valid reset gate position in the waveform. The determination of the reset gate position depends on the fluctuation of the echo signal and the environmental noise. The formula is as follows: T r =b·s window (i)+γ·Noise Where: T r is the adaptive threshold; σ window (i) is the standard deviation of the current window, indicating the fluctuation of the echo signal in the area; β and γ are adjustment parameters; Noise is the noise level, indicating the influence of environmental noise; when the signal exceeds the threshold, the point is determined as the re-gate position.

6. The method for monitoring glacial lake morphology in data-deficient areas based on satellite remote sensing according to claim 1, characterized in that: The atmospheric correction in step 3) uses the Sen2Cor and / or LEDAPS methods based on the radiation transfer model; the geometric correction uses Google Earth high-resolution remote sensing images as a reference for precise alignment to ensure the consistency of multi-source data in spatial positioning.

7. The method for monitoring glacial lake morphology in data-deficient areas based on satellite remote sensing according to claim 1, characterized in that: The calculation formulas for each water body index in step 5) are as follows: AWEI sh =4(Green-SWIR1) / (0.25NIR+2.75SWIR2) AWEI nsh =Blue+2.5Green-1.5(NIR+SWIR1)-0.25SWIR2 Among them: Green represents the green light band of the multispectral image, NIR represents the near-infrared band, Blue represents the blue light band, and SWIR represents the shortwave infrared band, which includes two bands SWIR1 and SWIR2, and the wavelength ranges are approximately 1550-1750nm and 2080-2350nm, respectively.

8. The method for monitoring glacial lake morphology in data-deficient areas based on satellite remote sensing according to claim 1, characterized in that: Step 5) The prediction of each decision tree in the random forest classification method is expressed as a function based on the four water index characteristics. The specific formula is as follows: Where: X j =[NDWI, MNDWI, AWEI sh , AWEI nsh ] are the four water index features input, R i is the i-th leaf node area, I(X j ∈R i ) is the indicator function, which represents the eigenvector X j Whether it belongs to the i-th leaf node, y i is the class label of the leaf node.

Images