Glacier surface accumulated snow identification method by using time sequence optics and SAR (Synthetic Aperture Radar) data

Abstract

The invention discloses a glacier surface accumulated snow identification method using time sequence optics and SAR data. The method comprises the following steps: generating a monthly-scale cloudless NDSI composite image; constructing a wet snow-non-wet snow binary training sample set of spatial-temporal distribution, and constructing a monthly-scale wet snow distribution prediction model based on a convolutional neural network to perform spatial-temporal feature learning; obtaining a month-by-month wet snow distribution diagram of a corresponding year, and calculating and generating a glacier surface annual wet snow coverage frequency distribution diagram; in combination with backscattering coefficient characteristics of the time sequence SAR image, establishing an SAR image wet snow discrimination threshold dynamic calculation model by counting backscattering coefficient distribution of the time sequence SAR image, and extracting wet snow distribution of the multi-temporal SAR image; and for the obtained residual area outside the wet snow distribution area, terrain slope direction parameters are extracted through a digital elevation model, classification constraints are constructed in combination with accumulated snow seasonal characteristics, and fine distinguishing of dry snow and glacier ice is realized. According to the invention, the automation degree and classification precision of high mountain area accumulated snow classification are improved.

Description

Technical Field

[0001] This invention relates to the application of synthetic aperture radar imagery in cryosphere remote sensing, specifically to a method for identifying snow accumulation on glacier surfaces using time-series optical and SAR data. Background Technology

[0002] Mountain glaciers, due to their sensitivity to climate change, are considered key indicators of global climate research. The Tibetan Plateau, as one of the regions with the highest concentration of mountain glaciers globally, provides important insights into regional hydrological cycles, global sea-level monitoring, and global climate change research through its glacier ablation processes. The surface cover type of mountain glaciers transitions between glacial ice and snow cover depending on the season; the latter can be further divided into wet snow and dry snow based on water content. An increase in wet snow distribution indicates melting of the glacier surface and can therefore be used as a basis for judging the glacier's ablation status.

[0003] However, glaciers are often located in extreme climate zones. Traditional meteorological station monitoring and manual field surveys are not only time-consuming and labor-intensive, but also cannot cover large areas, making it difficult to achieve effective long-term monitoring of glacier surface conditions. In recent years, optical remote sensing has made significant progress in glacier surface contour extraction, meltwater lake monitoring, and flow velocity estimation by taking advantage of the differences in reflectivity of different land features in glacier areas by spectral bands (Williamson AG, Banwell AF, Willis IC, et al. Dual-satellite (Sentinel-2 and Landsat 8) remote sensing of supraglacial lakes in Greenland[J].The Cryosphere,2018,12(9):3045-3065.). Some scholars have proposed the Normalized Difference Snow Index (NDSI) based on Landsat™ data, pointing out that snow has high reflectivity in the visible and near-infrared bands, but low reflectivity in the short-wave near-infrared band, thus enabling effective extraction of snow cover (Dozier J. Spectral signature of alpine snow cover from the Landsat Thematic Mapper[J]. Remote sensing of environment,1989,28:9-22.). However, traditional optical remote sensing is easily obscured by clouds and fog, resulting in poor spatiotemporal continuity. This can lead to discontinuities in long-term, high-density glacier ablation studies, and the inconsistent data quality makes the screening process costly in terms of manpower and resources.

[0004] SAR images have all-weather, all-day imaging capabilities and are highly sensitive to the moisture content in glaciers and snow. As the liquid water content in snow increases, the imaginary part of the dielectric constant of the snow increases significantly, resulting in a decrease in the backscattering coefficient (Ulaby FT, Stiles WH, Brunfeldt D, et al. 1-35GHz microwave scatterometer[C]. IEEE MTT-S International Microwave Symposium Digest, 1979: 551-553.). Utilizing this characteristic, by setting a backscattering threshold, SAR images can be used to classify dry and wet snow (Nagler T, Rott H, Ripper E, et al. Advancements for snowmelt monitoring by means of Sentinel-1SAR[J]. Remote Sensing, 2016, 8(4): 348.). Some scholars have proposed using Sentinel-1 time-series SAR images to extract wet snow pixels by comparing reference images with multi-temporal snowmelt images and combining backscattering of VV and VH polarizations with weights. They also propose extracting dry snow by combining the Sigmoid function with the average elevation of wet snow (Zhang Yanli, Chen Gang, Ma Yalong, et al. Study on surface ablation changes of mountain glaciers based on Sentinel-1 time-series data—taking Laohugou No. 12 Glacier as an example [J]. Acta Ecologica Sinica, 2024, 44(04): 1389-1403.). However, the threshold determination in this method relies on human experience and lacks reliability and applicability when widely promoted. Summary of the Invention

[0005] The purpose of this invention is to propose a method for identifying snow accumulation on the surface of glaciers using time-series optical and SAR data.

[0006] The technical solution to achieve the objective of this invention is: a method for identifying glacier surface snow accumulation using time-series optical and SAR data, comprising the following steps:

[0007] Step 1, Generating monthly composite normalized snow cover index image:

[0008] Acquire multispectral remote sensing image sets for each month of the target area over many years, calculate the normalized snow index for single-temporal multispectral remote sensing images, and generate monthly cloudless NDSI composite images through a pixel-level temporal median synthesis algorithm.

[0009] Step 2, Construction of optical image wet snow distribution prediction model:

[0010] Wet snow is labeled on the monthly cloudless NDSI synthetic image based on ground-measured wet snow distribution data. A spatially and temporally distributed wet snow-non-wet snow binarized training sample set is constructed. A monthly-scale wet snow distribution prediction model based on convolutional neural network is constructed to learn the spatial and temporal features.

[0011] Step 3, Calculation of annual wet snow cover frequency on glacier surface:

[0012] Input the monthly cloudless NDSI composite image of the year to be identified into the monthly wet snow distribution prediction model to obtain the monthly wet snow distribution map of the corresponding year. Calculate the frequency of wet snow occurrence pixel by pixel to generate the annual wet snow cover frequency distribution map of the glacier surface.

[0013] Step 4, Construction of SAR image wet snow adaptive discrimination model:

[0014] Based on the obtained annual wet snow cover frequency distribution map of the glacier surface, combined with the backscattering coefficient characteristics of time-series SAR images, a dynamic calculation model for wet snow discrimination threshold of SAR images is established by statistically analyzing the backscattering coefficient distribution of time-series SAR images, and the wet snow distribution of multi-temporal SAR images is extracted.

[0015] Step 5, Collaborative classification of all snow cover types:

[0016] For the remaining areas outside the obtained wet snow distribution area, topographic aspect parameters are extracted through digital elevation model, and classification constraints are constructed in combination with the seasonal characteristics of snow cover to achieve a fine distinction between dry snow and glacial ice. Finally, a full-element classification map containing wet snow, dry snow and glacial ice is output.

[0017] Further, in step 2, the optical image wet snow distribution prediction model is constructed, and the specific method is as follows:

[0018] Step 2.1: Select a certain year and determine the NDSI threshold based on the measured wet snow distribution data on the ground. This threshold is used to distinguish wet snow in the monthly NDSI images. Based on the determined NDSI threshold, process the monthly NDSI images of the 12 months of the year into a wet snow-non-wet snow binarized label training sample set. The training sample set contains monthly NDSI images with 12 channels, and each channel corresponds to the wet snow-non-wet snow binarized label image for each month.

[0019] Step 2.2: Stack the monthly cloudless NDSI composite images of December to form a 12-channel monthly composite NDSI image as input to the convolutional neural network, and use the wet snow-non-wet snow binarized label image as output to train the monthly wet snow distribution prediction model.

[0020] Further, in step 3, the annual wet snow cover frequency of the glacier surface is calculated using the following method:

[0021] Step 3.1: Using the trained monthly scale wet snow distribution prediction model, and taking the 12-channel monthly scale NDSI composite images of other years as input, predict the wet snow-non-wet snow binarized label images for the corresponding year.

[0022] Step 3.2: Based on the obtained wet snow-non-wet snow binarized label images, calculate the wet snow frequency distribution map. The annual wet snow frequency distribution is represented as follows:

[0023]

[0024] Where M wet M represents the number of months in which wet snow appears in the pixel. total W represents the total number of months. F This represents the frequency of wet snow occurrence in a pixel. By iterating through all pixels, we obtain the annual wet snow cover frequency distribution map of the glacier surface.

[0025] Further, in step 4, the adaptive discrimination model for wet snow in SAR images is constructed, and the specific method is as follows:

[0026] Step 4.1: Obtain the annual wet snow cover frequency distribution map of the glacier surface. For all SAR images of the study area in the same year, sort the backscatter values ​​of the SAR image at each pixel location according to intensity.

[0027] Step 4.2: Based on the wet snow frequency and SAR image count of the pixel in that year, obtain the SAR wet snow discrimination threshold location:

[0028] i = [W F ×N SAR (2)

[0029] In the formula, N SAR is the number of SAR images for that year, i is the position of the SAR wet snow discrimination threshold in the sorted SAR backscattering coefficient sequence, and [·] is the rounding operator;

[0030] Step 4.3: Obtain the SAR wet snow discrimination threshold based on the SAR wet snow discrimination threshold location:

[0031]

[0032] In the formula, T W To obtain the wet snow discrimination threshold for the SAR image;

[0033] Step 4.4: Apply the wet snow discrimination threshold of SAR images to the SAR images of the whole year to obtain the wet snow-non-wet snow binary map of SAR images for all periods of the year.

[0034] Further, in step 5, the collaborative classification of all snow cover types is performed using the following method:

[0035] Step 5.1: For the residual areas outside the wet snow distribution area, calculate the topographic aspect parameter α based on the DEM, establish a primary classification rule, and further classify dry snow and glacial ice.

[0036]

[0037] Step 5.2: Construct the snow season constraints within the summer time window T. summer In the inner months, priority is given to identifying glaciers with no snow cover, while in other months, priority is given to identifying them as having dry snow.

[0038]

[0039] Step 5.3: Combining topographic aspect parameters and seasonal parameters, and based on the location and specific type of the glacier, optimize the dry snow / glacier ice classification results to generate the final full-element classification map of wet snow, dry snow, and glacier ice.

[0040]

[0041] A glacier surface snow accumulation classification system based on multi-source remote sensing data is proposed. The system implements the glacier surface snow accumulation classification method based on multi-source remote sensing data to achieve the classification of glacier surface snow accumulation types based on multi-source remote sensing data. It consists of five modules, each executing steps 1 to 5.

[0042] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the glacier surface snow accumulation classification method based on multi-source remote sensing data, thereby achieving glacier surface snow accumulation classification based on multi-source remote sensing data.

[0043] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for classifying all types of glacier surface snow accumulation based on multi-source remote sensing data is implemented, thereby achieving classification of all types of glacier surface snow accumulation based on multi-source remote sensing data.

[0044] Compared with existing technologies, the present invention has the following significant advantages: 1) The wet snow distribution prediction model based on monthly cloudless NDSI synthetic images and CNN effectively avoids data loss caused by cloud and fog obstruction in optical images, and realizes automated monthly wet snow-non-wet snow discrimination in optical images, which has strong practicality; 2) The SAR wet snow adaptive discrimination model based on optical wet snow frequency map transfers the wet snow discrimination capability of optical images to adaptively generate the wet snow discrimination threshold of SAR images, breaking through the limitation of setting empirical thresholds in traditional methods, and realizing automated wet snow discrimination based on SAR images; 3) By combining topographic and seasonal constraints, the SAR image time series can achieve fine classification of all snow cover types, while being unaffected by weather factors, and can provide a more dense glacier surface snow cover category product than optical remote sensing. Attached Figure Description

[0045] Figure 1 This document outlines the overall technical process for a method of identifying snow accumulation on glacier surfaces using time-series optical and SAR data.

[0046] Figure 2 A classification process for all types of surface snow accumulation on glaciers in the Qinghai-Tibet Plateau based on Sentinel-1 and Sentinel-2.

[0047] Figure 3 The annual wet snow frequency for the three glaciers.

[0048] Figure 4 The classification results for the glacier ablation period in 2021 are as follows: (a) Shenshechuan Glacier; (b) Laohugou No. 12 Glacier; (c) Bayi Glacier.

[0049] Figure 5 The temporal variation of the ratio of dry to wet snow on different glaciers. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0051] A method for identifying glacier surface snow accumulation using time-series optical and SAR data includes the following steps:

[0052] Step 1, Generating monthly composite normalized snow index (NDSI) imagery:

[0053] The monthly multispectral remote sensing image set of the target area was acquired, the normalized snow index (NDSI) was calculated on the single-temporal multispectral remote sensing images, and a monthly cloudless NDSI composite image was generated by the pixel-level temporal median synthesis algorithm.

[0054] Step 2, Construction of optical image wet snow distribution prediction model:

[0055] Based on ground-measured wet snow distribution data, wet snow is labeled on monthly NDSI images. A spatiotemporally distributed binary training sample set of wet and non-wet snow is constructed. Spatiotemporal features are learned through a convolutional neural network (CNN) to obtain a trained monthly-scale wet snow distribution prediction model. The specific method is as follows:

[0056] Step 2.1: Select a certain year and determine the NDSI threshold based on the measured wet snow distribution data on the ground. This threshold is used to distinguish wet snow in the monthly NDSI image. Based on the threshold determined by NDSI, the monthly NDSI images of the 12 months of the year are processed into a wet snow-non-wet snow binarized label training sample set. The training sample set contains monthly NDSI images with 12 channels, and each channel corresponds to the wet snow-non-wet snow binarized label image of each month.

[0057] Step 2.2 involves stacking the monthly cloudless NDSI composite images from December to form a 12-channel monthly NDSI image as input, and using the wet snow-non-wet snow binarized label images as output. This process fully trains the CNN model as a monthly wet snow distribution prediction model, enabling the model to distinguish between wet and non-wet snow.

[0058] Step 3, Calculation of annual wet snow cover frequency on glacier surface:

[0059] The monthly-scale NDSI images from historical years are input into the monthly-scale wet snow distribution prediction model to obtain the monthly wet snow distribution map for the corresponding year. The frequency of wet snow occurrence is calculated pixel by pixel through time-series overlay to generate the annual wet snow cover frequency distribution map of the glacier surface. The specific method is as follows:

[0060] Step 3.1: Using the trained monthly wet snow distribution prediction model, and taking the 12-channel monthly composite NDSI image of other years as input, predict the wet snow-non-wet snow binarized label image of the corresponding year.

[0061] Step 3.2: Based on the obtained monthly wet snow-non-wet snow binarized label images, calculate the wet snow frequency distribution map. The annual wet snow frequency distribution can be calculated using the following formula:

[0062]

[0063] Where M wet M represents the number of months in which wet snow appears in the pixel. total This represents the total number of months; if there are no missing data, it is generally 12. W F This represents the frequency of wet snow occurrence in a pixel. By using this formula to iterate through all pixels, we can obtain the annual wet snow cover frequency distribution map of the glacier surface.

[0064] Step 4, Construction of SAR image wet snow adaptive discrimination model:

[0065] Based on the obtained annual wet snow cover frequency distribution map of the glacier surface, and combined with the backscattering coefficient characteristics of time-series SAR images, a dynamic calculation model for the wet snow discrimination threshold of SAR images is established by statistically analyzing the backscattering coefficient distribution of time-series SAR images. This enables automatic extraction of wet snow distribution from multi-temporal SAR images. The specific method is as follows:

[0066] Step 4.1: Obtain the annual wet snow cover frequency distribution map of the glacier surface. For all SAR images of the study area in the same year, sort the backscatter values ​​of the SAR image at each pixel location according to intensity.

[0067] Step 4.2: Based on the wet snow frequency and SAR image count of the pixel in that year, automatically obtain the SAR wet snow discrimination threshold location:

[0068] i = [W F ×N SAR In equation (2), N SAR denoted as the number of SAR images for that year, i represents the position of the SAR wet snow discrimination threshold in the sorted SAR backscattering coefficient sequence, and [·] represents the rounding operator.

[0069] Step 4.3: Obtain the SAR wet snow discrimination threshold based on the SAR wet snow discrimination threshold location:

[0070]

[0071] In the formula, T W The threshold for wet snow discrimination in the obtained SAR image.

[0072] Step 4.4: Apply the wet snow discrimination threshold of SAR images to the SAR images of the whole year to obtain the wet snow-non-wet snow binary map of SAR images for all periods of the year.

[0073] Step 5, Collaborative classification of all snow cover types:

[0074] For the residual areas outside the obtained wet snow distribution area, topographic aspect parameters are extracted using a digital elevation model (DEM). Classification constraints are constructed by combining these parameters with seasonal snow cover characteristics to achieve a refined distinction between dry snow and glacial ice. The final output is a comprehensive classification map containing wet snow, dry snow, and glacial ice. The specific method is as follows:

[0075] Step 5.1: For the residual area outside the wet snow distribution area, calculate the topographic aspect parameter α based on DEM. According to α, divide the terrain into shady slope area α∈[135°,225°] and sunny slope area α∈[0°,45°]∪[315°,360°]. Establish a primary classification rule to further classify dry snow and glacial ice.

[0076]

[0077] Step 5.2, construct the snow season constraints, within the summer time window (T summer ( ), priority is given to identifying glaciers with no snow cover, while in other months they are identified as having dry snow;

[0078]

[0079] Step 5.3: Combining topographic aspect parameters and seasonal parameters, and based on the location and specific type of the glacier, optimize the dry snow / glacier ice classification results to generate the final full-element classification map of wet snow, dry snow, and glacier ice.

[0080]

[0081] This invention also proposes a glacier surface snow accumulation type classification system based on multi-source remote sensing data. The system implements the glacier surface snow accumulation type classification method based on multi-source remote sensing data to achieve glacier surface snow accumulation type classification based on multi-source remote sensing data. It consists of five modules, each executing steps 1 to 5.

[0082] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the glacier surface snow accumulation classification method based on multi-source remote sensing data, thereby achieving glacier surface snow accumulation classification based on multi-source remote sensing data.

[0083] A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for classifying all types of glacier surface snow accumulation based on multi-source remote sensing data is implemented, thereby achieving classification of all types of glacier surface snow accumulation based on multi-source remote sensing data.

[0084] Example

[0085] To verify the effectiveness of the present invention, the following experiment was conducted.

[0086] This invention provides a method for identifying snow accumulation on glaciers using time-series optical and SAR data. The optical data utilizes Sentinel-2, a high-resolution multispectral optical satellite from the European Space Agency's Copernicus program, with 13 bands (covering visible to shortwave infrared), a spatial resolution of 10-60 meters, and a global revisit period of 5 days after the two satellites are networked. The SAR data utilizes Sentinel-1, which has all-weather, all-day imaging capabilities and can identify snow accumulation and glacier surface features by analyzing backscattering coefficients to penetrate cloud layers. The two data sets are complementary. The specific implementation process is as follows... Figure 2The flowchart shown is a classification of all types of surface snow accumulation on glaciers in the Tibetan Plateau based on Sentinel-1 and Sentinel-2, including steps 1 to 5 below.

[0087] Step 1: Acquire time-series Sentinel-2 images and calculate monthly cloud-free NDSI composite images for all years. The specific steps are as follows:

[0088] Step 1.1: Obtain the Sentinel-2 surface reflectance time series images with a spatial resolution of 10 meters from 2019 to 2024. Calculate the NDSI using the green surface reflectance (B03) and shortwave infrared surface reflectance (B11) from these images. The calculation formula is as follows:

[0089]

[0090] Step 1.2: For NDSI images of a single time phase within the month, the NDSI value of each pixel is counted, the median is calculated, and all NDSI images within the month are combined into a monthly cloudless NDSI composite image. Considering that Sentinel-2 has a relatively high revisit period, the monthly composite image generally does not have any missing data. Therefore, there are a total of 12 monthly cloudless NDSI composite images throughout the year.

[0091] Step 2: Using the obtained Sentinel-2 month-scale cloudless NDSI composite image, select one year as the training sample to train a CNN model for inferring the wet snow distribution on Sentinel-2 month-scale cloudless NDSI composite images of other years. Specific steps include:

[0092] Step 2.1: Using 2024 as the training sample, determine the NDSI threshold based on the ground measurement data of 2024. Use this threshold to process the monthly cloudless NDSI composite images of each month in 2024 into a wet snow-non-wet snow binarized label training sample set, and combine the sample set into a monthly NDSI image with 12 channels, each channel corresponding to a wet snow-non-wet snow binarized label image for one month.

[0093] Step 2.2: Stack the monthly cloudless NDSI composite images of the 12 months of 2024 to form a 12-channel monthly composite NDSI image as input, and use the wet snow-non-wet snow binarized label image as output to train the CNN model so that the model has the ability to distinguish wet snow.

[0094] Step 3: The CNN model trained based on the 2024 Sentinel-February monthly cloud-free NDSI synthetic image can be used to infer the distribution of wet snow in other historical years (2019-2023), and then calculate the annual wet snow cover frequency distribution map of the glacier surface. The specific steps include:

[0095] Step 3.1: Using the monthly scale wet snow distribution prediction model trained by Sentinel-2 in 2024, and taking the monthly synthetic NDSI images of other years (2019-2023) as input, predict the wet snow-non-wet snow binarized label images for the corresponding years.

[0096] Step 3.2: Based on the obtained 2019-2023 wet snow-non-wet snow binarized label images, calculate the wet snow frequency distribution map for each year. The annual wet snow frequency distribution can be calculated using the following formula:

[0097]

[0098] Where M wet M represents the number of months in which wet snow appears in the pixel. total This represents the total number of months, which is 12 for these years. F This represents the frequency of wet snow occurrence in a pixel. By iterating through all pixels using this formula, a distribution map of the annual wet snow cover frequency on the glacier surface can be obtained, as shown below. Figure 3 As shown. Figure 3 The data shows the annual wet snow frequency distribution of Bayi Glacier, Laohugou Glacier and Shenshechuan Glacier from 2019 to 2024, which is generally between 0 and 0.5.

[0099] Step 4: Using the annual wet snow cover frequency distribution map of the glacier surface obtained based on Sentinel-2, combined with the backscattering pattern of Sentinel-1, the wet snow discrimination threshold of the SAR image can be automatically obtained. The specific steps are as follows:

[0100] Step 4.1: Obtain the annual wet snow cover frequency distribution map of the glacier surface. For all SAR images of the study area in the same year, sort the backscatter values ​​of the SAR image at each pixel location according to the intensity. It is considered that the backscatter value of the pixel with wet snow is less than that of the dry snow and ice.

[0101] Step 4.2: Based on the Sentinel-2 wet snow frequency and the number of SAR images for that pixel in that year, automatically obtain the SAR wet snow discrimination threshold position, i.e., if there are N wet snow images in that year... SAR Scenery SAR imagery, with wet snow frequency at W. F Then the backscattering is less than W F ×N SAR All backscattered data can be considered as wet snow pixels. Therefore, the position i of the wet snow discrimination threshold in the sorted SAR backscattering coefficient sequence can be expressed as:

[0102] i = [W F ×N SAR ]

[0103] In the formula, [·] is the floor operator.

[0104] Step 4.3: Based on the location of the SAR wet snow discrimination threshold, the wet snow discrimination threshold of the SAR image can be obtained from the SAR backscatter sequence.

[0105]

[0106] In the formula, T W The threshold for wet snow discrimination in the obtained SAR image.

[0107] Step 4.4: Apply the wet snow discrimination threshold of the SAR image to the Sentinel-1 SAR backscatter intensity image for each scene throughout the year to obtain the wet snow-non-wet snow binary map of the SAR image for all periods of the year. The Sentinel-1 wet snow-non-wet snow binary map obtained at this time is denser than the result of the optical image and is not affected by cloud and fog obstruction, enabling continuous observation of wet snow on the glacier surface.

[0108] Step 5: Combining DEM and seasonal constraints, further classify the non-wet snow areas, dividing the glacier surface type into three categories: wet snow, dry snow, and glacial ice. The specific steps are as follows:

[0109] Step 5.1: For the residual areas outside the wet snow distribution area, calculate the topographic aspect parameter α based on the DEM. Divide the terrain into shaded slope areas α∈[135°,225°] and sunny slope areas α∈[0°,45°]∪[315°,360°] based on α, and establish a primary classification rule:

[0110]

[0111] Step 5.2, construct the snow season constraints, within the summer time window (T summer The system prioritizes identifying glaciers with no snow cover, while other months are prioritized for identifying dry snow. In this embodiment, the summer afternoon time window is from June to August.

[0112]

[0113] Step 5.3: Combining topographic aspect parameters and seasonal parameters, and based on the location and specific type of the glacier, optimize the dry snow / glacier ice classification results to generate the final comprehensive classification map of wet snow, dry snow, and glacier ice.

[0114]

[0115] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0116] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these modifications and improvements all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for identifying glacier surface snow accumulation using time-series optical and SAR data, characterized in that, Includes the following steps: Step 1, Generating monthly composite normalized snow cover index image: Acquire multispectral remote sensing image sets for each month of the target area over many years, calculate the normalized snow index for single-temporal multispectral remote sensing images, and generate monthly cloudless NDSI composite images through a pixel-level temporal median synthesis algorithm. Step 2, Construction of optical image wet snow distribution prediction model: Wet snow is labeled on the monthly cloudless NDSI synthetic image based on ground-measured wet snow distribution data. A spatially and temporally distributed wet snow-non-wet snow binarized training sample set is constructed. A monthly-scale wet snow distribution prediction model based on convolutional neural network is constructed to learn the spatial and temporal features. Step 3, Calculation of annual wet snow cover frequency on glacier surface: Input the monthly cloudless NDSI composite image of the year to be identified into the monthly wet snow distribution prediction model to obtain the monthly wet snow distribution map of the corresponding year. Calculate the frequency of wet snow occurrence pixel by pixel to generate the annual wet snow cover frequency distribution map of the glacier surface. Step 4, Construction of SAR image wet snow adaptive discrimination model: Based on the obtained annual wet snow cover frequency distribution map of the glacier surface, combined with the backscattering coefficient characteristics of time-series SAR images, a dynamic calculation model for the wet snow discrimination threshold of SAR images is established by statistically analyzing the backscattering coefficient distribution of time-series SAR images, and the wet snow distribution of multi-temporal SAR images is extracted. Step 5, Collaborative classification of all snow cover types: For the remaining areas outside the obtained wet snow distribution area, topographic aspect parameters are extracted through digital elevation model, and classification constraints are constructed in combination with the seasonal characteristics of snow cover to achieve a fine distinction between dry snow and glacial ice. Finally, a full-element classification map containing wet snow, dry snow and glacial ice is output.

2. The method for classifying all types of glacier surface snow cover based on multi-source remote sensing data according to claim 1, characterized in that, Step 2, constructing a wet snow distribution prediction model from optical images, the specific method is as follows: Step 2.1: Select a certain year and determine the NDSI threshold based on the measured wet snow distribution data on the ground. This threshold is used to distinguish wet snow in the monthly NDSI images. Based on the determined NDSI threshold, process the monthly NDSI images of the 12 months of the year into a wet snow-non-wet snow binarized label training sample set. The training sample set contains monthly NDSI images with 12 channels, and each channel corresponds to the wet snow-non-wet snow binarized label image for each month. Step 2.2: Stack the monthly cloudless NDSI composite images of December to form a 12-channel monthly composite NDSI image as input to the convolutional neural network, and use the wet snow-non-wet snow binarized label image as output to train the monthly wet snow distribution prediction model.

3. The method for classifying all types of glacier surface snow cover based on multi-source remote sensing data according to claim 1, characterized in that, Step 3, Calculation of annual wet snow cover frequency on glacier surface, the specific method is as follows: Step 3.1: Using the trained monthly scale wet snow distribution prediction model, and taking the 12-channel monthly scale cloudless NDSI composite images of other years as input, predict the wet snow-non-wet snow binarized label images of the corresponding years. Step 3.2: Based on the obtained wet snow-non-wet snow binarized label images, calculate the wet snow frequency distribution map. The annual wet snow frequency distribution is represented as follows: Where M wet M represents the number of months in which wet snow appears in the pixel. total W represents the total number of months. F This represents the frequency of wet snow occurrence in a pixel. By iterating through all pixels, we obtain the annual wet snow cover frequency distribution map of the glacier surface.

4. The method for classifying all types of glacier surface snow cover based on multi-source remote sensing data according to claim 1, characterized in that, Step 4: Construction of the adaptive discrimination model for wet snow in SAR images. The specific method is as follows: Step 4.1: Obtain the annual wet snow cover frequency distribution map of the glacier surface. For all SAR images of the study area in the same year, sort the backscatter values ​​of the SAR image at each pixel location according to intensity. Step 4.2: Based on the wet snow frequency and SAR image count of the pixel in that year, obtain the SAR wet snow discrimination threshold location: i = [W F ×N SAR In equation (2), N SAR is the number of SAR images for that year, i is the position of the SAR wet snow discrimination threshold in the sorted SAR backscattering coefficient sequence, and [·] is the rounding operator; Step 4.3: Obtain the SAR wet snow discrimination threshold based on the SAR wet snow discrimination threshold location: In the formula, To obtain the wet snow discrimination threshold for the SAR image; Step 4.4: Apply the wet snow discrimination threshold of SAR images to the SAR images of the whole year to obtain the wet snow-non-wet snow binary map of SAR images for all periods of the year.

5. The method for classifying all types of glacier surface snow cover based on multi-source remote sensing data according to claim 1, characterized in that, Step 5, Collaborative classification of all snow cover types, the specific method is as follows: Step 5.1: For the residual areas outside the wet snow distribution area, calculate the topographic aspect parameter α based on the DEM, establish classification rules, and further divide dry snow and glacial ice. Step 5.2: Construct the snow season constraints within the summer time window T. summer In the inner months, priority is given to identifying glaciers with no snow cover, while in other months, priority is given to identifying them as having dry snow. Step 5.3: Combining topographic aspect parameters and seasonal parameters, and based on the location and specific type of the glacier, optimize the dry snow / glacier ice classification results to generate the final full-element classification map of wet snow, dry snow, and glacier ice.

6. A classification system for all types of glacier surface snow accumulation based on multi-source remote sensing data, characterized in that, The method for classifying all types of glacier surface snow accumulation based on multi-source remote sensing data according to any one of claims 1-5 is implemented to achieve the classification of all types of glacier surface snow accumulation based on multi-source remote sensing data. It is divided into five modules, and steps 1 to 5 are executed respectively.

7. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements the glacier surface snow accumulation classification method based on multi-source remote sensing data as described in any one of claims 1-5, thereby realizing the glacier surface snow accumulation classification based on multi-source remote sensing data.

8. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, it implements the glacier surface snow accumulation classification method based on multi-source remote sensing data as described in any one of claims 1-5, thereby realizing the classification of glacier surface snow accumulation types based on multi-source remote sensing data.

Images