Method and system for monitoring sea surface elevation of ice-covered sea area based on satellite height measurement data

Abstract

The invention provides an ice-covered sea area sea surface elevation monitoring method and system based on satellite height measurement data, and belongs to the technical field of sea surface elevation monitoring. Based on a plurality of characteristic parameters of SAR mode radar echo data, an unsupervised clustering method is adopted to carry out initial classification on waveforms, and then a gradient boosting tree model is combined to optimize a classification result, so that the sea surface elevation of an ice-covered sea area is obtained. Carrying out classification processing on the waveform; according to a waveform classification result, carrying out distance correction on different types of waveforms by adopting different empirical re-tracking algorithms; on the basis of the re-tracked distance data, calculating sea surface elevation data by combining a satellite orbit height, a geophysical correction item and a geoid model; projecting sea surface elevation data to an equal-area grid, performing spatial interpolation and smoothing processing on a missing region, and generating a monthly sea surface elevation data set covering the ice-covered sea area; and comparing and verifying the generated sea surface elevation data with tide station data and an existing data set, and outputting a final sea surface elevation monitoring result.

Description

Technical Field

[0001] This invention relates to the field of sea surface elevation monitoring technology, specifically to a method and system for monitoring sea surface elevation in ice-covered sea areas based on satellite altimetry data. Background Technology

[0002] Against the backdrop of global warming, the melting rate of Arctic sea ice is accelerating, which in turn contributes to changes in the Arctic Ocean's sea level. With increasing research, satellite altimetry has proven to be a crucial technique for calculating these changes. However, due to sea ice cover, the Arctic Ocean's sea level data cannot be directly measured, necessitating specialized techniques for data processing.

[0003] Sea level changes are often the most direct reflection of climate change. The Arctic Ocean is influenced by a combination of factors in the climate system and is extremely sensitive to climate change; its sea surface dynamics and topography reflect almost all of these changes. Therefore, by studying the changes in the Arctic Ocean's sea surface dynamics and topography, we can gain a deeper understanding of the key processes in the evolution of the Arctic Ocean and the Arctic climate system. Researchers have further expanded the latitude and time span of Arctic Ocean sea surface dynamic topography data by utilizing data from different satellite altimeters. Armitage et al. (2016) studied and provided 12 years of data for the Arctic Ocean from 2003 to 2014, and subsequently updated the data to 2020, with the northernmost spatial range reaching 81.5°N. Rosee et al. (2019) constructed a lunar sea level dataset for the Arctic Ocean from 60°N to 81.5°N using 27 years of radar satellite altimeter data. Prandie et al. (2021) fused multiple satellite altimeter data to obtain a large-scale continuous altimeter observation dataset for the Arctic Ocean, covering approximately 20 years of sea surface dynamic topography records, with a spatial range reaching all areas north of 50°N and a spatial resolution of 25km grid. Summary of the Invention

[0004] To address the aforementioned issues, this invention, based on the Cryosat-2 satellite, analyzed waveform classification and retracking methods previously used in research. New processing was applied to these methods according to the Arctic situation, resulting in a gridded DOT (dynamic oceantopography) dataset covering the Arctic Ocean and ice-covered sea areas between 60° and 85° north latitude, spanning from 2011 to 2024, with a spatial resolution of 25 km. This dataset was evaluated against published DOT datasets and SLA (Sea Level Anomaly) data from tide gauge stations over time, demonstrating the feasibility of the method and the usability of the data.

[0005] A method for monitoring sea surface elevation in ice-covered sea areas based on satellite altimetry data includes the following steps: S1. Acquire SAR mode radar echo data from the CryoSat-2 satellite in ice-covered sea areas, and simultaneously acquire sea ice concentration data and optical remote sensing image data. S2. Based on multiple feature parameters of SAR mode radar echo data, an unsupervised clustering method is used to initially classify the waveforms, and then the classification results are optimized by combining a gradient boosting tree model to classify the waveforms. S3. Based on the waveform classification results, different empirical re-tracking algorithms are used for distance correction for different types of waveforms; S4. Based on the distance data after retracking, combined with satellite orbital altitude, geophysical correction terms and geoid model, calculate sea surface elevation data; S5. Project the sea surface elevation data onto an equal-area grid, perform spatial interpolation and smoothing on the missing areas, and generate a monthly sea surface elevation dataset covering ice-covered sea areas. S6. Compare and verify the generated sea surface elevation data with the tide gauge station data and the existing dataset, and output the final sea surface elevation monitoring results.

[0006] Furthermore, in step S2, the K-medoids clustering algorithm is used to perform unsupervised initial classification of radar echo waveforms. Based on the clustering results, the waveforms are categorized into four major categories: ocean waveforms, interglacial water waveforms, sea ice waveforms, and undefined waveforms. The categorized results are then used as training data and input into the gradient boosting tree model for supervised classification.

[0007] Furthermore, in step S3, the maximum threshold re-tracking algorithm is used for the interglacial water waveform to locate the re-tracking point where the waveform first rises rapidly; the main peak threshold re-tracking algorithm is used for the ocean waveform to locate the position of the main peak of the waveform.

[0008] Furthermore, the main peak threshold re-tracking algorithm calculates the thresholds at the start and end of the peak. and This is used to locate the main peak of the waveform; Calculate the power difference between two sampling points that are one sampling point apart. : ; Calculate the threshold at the start of the peak : ; Calculate the power difference between two adjacent waveform sampling points : ; Calculate the threshold at the end of the peak : ; Where N is the total number of sampling points of the waveform to be calculated. This represents the power at the location of the i-th waveform sampling point. This represents the power at the (i+1)th waveform sampling point.

[0009] Furthermore, the sea surface elevation data includes sea surface height (SSH), sea surface dynamic topography (DOT), and absolute dynamic topography (ADT); ; ; ; Where Orbit is the satellite's orbital altitude relative to the WGS84 ellipsoid; Range is the distance from the satellite to the ground surface after retracking calculation; It is a tidal and atmospheric geophysical correction; It represents the height of the geoid; SLA represents the sea level height anomaly; MDT represents the mean dynamic topography.

[0010] Furthermore, the tidal and geophysical corrections include: orbital correction, ionospheric correction, wet tropospheric correction, dry tropospheric correction, dynamic atmospheric correction, solid tide, ocean tide, and polar tide correction.

[0011] Furthermore, the sea level height anomaly SLA is: ; MSS represents the mean sea level.

[0012] Furthermore, the equal-area grid in step S5 is an EASE2 projection grid with a resolution of 25km; the spatial interpolation is Kriging interpolation; and the smoothing process is Gaussian smoothing at a scale of 100km.

[0013] Furthermore, the data verification in step S6 includes comparing the generated sea surface elevation data with the tide gauge data, CPOM dataset and / or PADT dataset provided by PSMSL, and evaluating the data accuracy by calculating the correlation coefficient and root mean square deviation.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This technology improves the classification accuracy of radar echoes from the Arctic Ocean. Compared to traditional thresholding methods, it eliminates the need for pre-summarizing threshold ranges, reducing misclassification. Furthermore, the strict numerical range of thresholding methods often leads to a large amount of data remaining unclassified; this technique addresses this issue. In contrast to using machine learning methods alone, which rely on parameters reflecting the mathematical characteristics of waveforms as samples, many waveforms remain difficult to define. This technique first selects six mathematical feature parameters of the waveform for unsupervised classification, combined with visual interpretation. This results in a more robust model during subsequent training, enhancing the recognition rate of complex waveforms and allowing for increased data volume for subsequent computations.

[0015] Different re-tracking algorithms are employed based on the characteristics of different waveforms to improve computational accuracy. For ocean waveforms, an algorithm that extracts the main peak is used, which differs from the computational focus of other traditional ocean waveforms. Instead, the computational focus is placed on the main peak, reducing the impact of other possible peaks. Furthermore, the algorithm is improved based on the latest Baseline-E version from the Cryosat-2 satellite. Attached Figure Description

[0016] Figure 1 A schematic diagram of the echo waveforms of the CryoSat-2 satellite on different surface types; Figure 2 This is a graph showing the results after clustering using K-medoids; Figure 3 A precision map of the confusion matrix for the classification model; Figure 4 CryoSat-2 footprints were superimposed on the MODISMOD02QKM image from April 2014. Figure 5 A schematic diagram of the algorithms for TFMRA and PPTRA; Figure 6 This is a monthly map showing the dynamic distribution of Arctic ocean topography in March of each year from 2011 to 2024. Figure 7 (a) shows the difference in mean sea surface dynamic topography between the CryoSat-2 and CPOM datasets; (b) shows the variance ratio of sea surface dynamic topography. Figure 8 (a) Mean absolute dynamic terrain difference between CryoSat-2 data and PADT dataset; (b) Variance ratio of absolute dynamic terrain. Figure 9 This is the SLA time series data of sea surface height anomalies from the three tide gauge stations: Kigliah, Stellova, and Alert. Detailed Implementation

[0017] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0018] 1. Data 1.1 Altimeter Radar The CryoSat-2 radar altimeter satellite covers a vast area of ​​the Arctic Ocean. With an along-orbit spatial resolution of 250-300 meters, it can accurately identify cracks and interglacial water within sea ice-covered areas, thus providing more precise altimetry data. CryoSat-2 has an orbital period of 369 days, consisting of continuously varying 30-day sub-cycles. This repeating pattern ensures extremely dense orbital coverage of the Arctic Ocean. Therefore, CryoSat-2 possesses the capability for high-frequency, repeated observations of the Arctic Ocean, which is one of the important reasons why it is a valuable source of research data.

[0019] CryoSat-2 L1B level data is divided into three modes. This invention uses CS2L1B level SAR mode data from 2010 to 2024. The data types are: LTA (Long Term Archive) from January 2011 to August 2021, and OFFL (Offline Processing / Systematic) from September 2021 to December 2024.

[0020] In SAR mode, each strip footprint is approximately 250 meters wide along the orbital direction and spans up to 15 kilometers across the orbit. The interval between satellite pulse bursts is designed to be 250 meters, ensuring that each ground location is observed multiple times. The strips formed by the continuous pulse arrangement can be superimposed and averaged to reduce noise. The basic parameters of CryoSat-2 are shown in Table 1.

[0021] Table 1. Basic parameters of the CryoSat-2 altimeter satellite

[0022] 1.2 Sea ice concentration This invention utilizes the OSI-450 (Leverage et al. 2019) and OSI-458 sea ice concentration products published by OSISAF (Ocean and Sea Ice Satellite Application Facilities) (OSI-450 data covers 2011-2020, and OSI-458 data covers 2021-2024). These products provide sea ice concentration data in a 10-kilometer resolution polar stereo grid. They offer high temporal and spatial resolution, providing accurate supplementation to field observation data across coastal and Arctic regions. This invention uses them to distinguish sea ice regions and assist in verifying the classification of interglacial water.

[0023] 1.3 Image Data Because of the significant difference in reflectivity between interglacial water and sea ice, using optical remote sensing imagery is an ideal method for separating interglacial channels and sea ice. This invention uses NASA's MOD02QKM data, with a spatial resolution of 250 meters, comparable to CryoSat-2, making it highly suitable as validation data. On MODIS imagery, numerous narrow black shadows can be seen in the sea ice region; these shadows represent interglacial water.

[0024] 1.4 Comparison Data Currently, the only published fully processed sea surface dynamic topography data covering Arctic ice-covered areas are Armitage et al. (2016)'s CPOM (data updated to 2020) and Prandie et al. (2021)'s DTARCECTSALTO / DUACSL4SeaSurfaceHeightProduct (hereinafter referred to as PADT), both of which are ten years or more in length. In addition, tide gauge data is also an important data source for accuracy assessment. However, tide gauges are sparsely distributed in the Arctic region. To ensure objectivity in comparison, this invention selected 15 tide gauge stations at different latitudes and in different regions published by PSMSL (Permanent Service for Mean Sea Level), which have data records from 2011 onwards.

[0025] 2. Methods 2.1 Classification of Oceanic and Interglacial Water The Arctic region is extensively covered by sea ice, with high-latitude areas frozen year-round, making it impossible to calculate sea surface dynamics and topography by directly measuring sea level. Interglacial water, formed by the horizontal movement of sea ice, is a typical feature of the Arctic Ocean and a crucial data source for calculating Arctic sea level. In the Arctic Ocean, interglacial water and newly formed thin ice surfaces are flat and smooth, exhibiting strong specular reflection properties, resulting in distinct peaks in their radar echo waveforms. In contrast, the surfaces of seawater and sea ice are relatively rough, leading to diffuse reflection of radar signals. Therefore, analyzing the waveform characteristics of radar echoes can effectively classify different surface types in the Arctic Ocean.

[0026] Figure 1 The following are schematic diagrams of echo waveforms from the CryoSat-2 satellite on different surface types: (a) typical ocean ("Brown") waveform, (b) interglacial channel waveform, and (c) sea ice waveform. Figure 1 The waveforms of typical Arctic Ocean features fall into three categories. The first category is the Brownian waveform, which has the largest waveform width and the slowest decay at the trailing edge, and is usually classified as an ocean waveform. The second category has waveform energy that reaches its maximum value rapidly and decays quickly, which is consistent with the specular reflection characteristics of interglacial water. The third category has a waveform width between the first and third categories and is classified as a sea ice waveform. The remaining waveforms, which are complex in shape and difficult to classify, are collectively referred to as unclassified waveforms.

[0027] A combination of unsupervised classification and gradient boosting trees was chosen to classify radar waveforms in the Arctic Ocean, improving classification accuracy and model applicability. The first step in unsupervised classification is to analyze the characteristics of the echo waveforms, requiring the selection of feature parameters that can describe the waveform differences between different surface types. The six parameters selected in this study include mathematical characteristics of the waveform such as pulse kurtosis (PP), trailing edge to leading edge ratio (ETPP), stacking standard deviation (SSD), and leading edge width (LEW), as well as sea ice concentration (SIC) and radar backscattering coefficient (Sigma0), which reflect the physical significance at the satellite footpoint.

[0028] (1) Pulse Peak PP: This is the ratio of the maximum energy value in the echo waveform to the average energy value of all echoes. This parameter describes the sharpness of the waveform, and the calculation formula is: ; (2) Early Tail to Peak Power Ratio (ETPP): Defined as the average echo energy in a certain range before the falling edge of the pulse return waveform divided by the maximum echo energy, as shown below. The commonly used range is from (max+1) to (max+6) as the falling edge for calculation. The formula is as follows: ; (3) Stack Standard Deviation (SSD): The standard deviation corresponding to the leading edge of the echo waveform after fitting a Gaussian function. The calculation formula is: ; (3) Leading Edge Width (LEW): This represents the difference between the starting point of the pulse return waveform and the point of maximum echo energy, typically 1% of the waveform leading edge amplitude. The widths of the bins corresponding to 9% are as follows: ; Sea ice concentrations (SIC) were extracted from OSI450 and OSI458 data.

[0029] The radar backscatter coefficient Sigma0 reflects the strength of radar echo energy, and the formula is as follows: ; ; First, these parameters were selected as the initial classification criteria for the waveforms. Using a threshold method, 50,000 waveforms each of four categories—seawater, interglacial water, sea ice, and undefined waveforms—were selected as samples for the waveform clustering algorithm. Then, the K-medoids clustering algorithm was used for unsupervised classification of the waveforms. The cluster size (K) was chosen to be 20. Since only four basic waveform categories are needed in the Arctic, the choice of 20 effectively represents intra-cluster differences while avoiding over-subdivision. This value was achieved through empirical testing, quantitative evaluation, and visual inspection. The K-medoids clustering algorithm performed better than the K-means clustering algorithm because K-medoids selects actual points in the dataset, unaffected by mean calculation. It also uses a more robust distance metric, reducing the impact of outliers and exhibiting stronger resistance to noise. The clustering results are shown below. Figure 2 The results of K-medoids clustering (K=20) show twenty waveforms for each class.

[0030] Based on the clustering results, classes 3, 14, and 19 can be classified as class 1 ocean waveforms; classes 1, 11, 16, and 17 can be classified as class 2 interglacial water waveforms; classes 2, 4, 5, 7, 8, 10, 12, 13, 15, and 18 can be classified as class 3 sea ice waveforms; and classes 9, 6, and 20 are classified as undefined waveforms. After dividing the clustered data into four waveform classes, it is used as training data for the gradient boosting tree algorithm. The reason for choosing unsupervised classification combined with gradient boosting trees is that only parameters describing the waveform shape features are used as input sample data for the gradient boosting tree. Using the results of unsupervised classification as input to the gradient boosting tree can improve robustness to complex waveforms and increase the amount of data for subsequent DOT calculations.

[0031] Figure 3 This is a confusion matrix accuracy plot of the classification model. The model exhibits high consistency between the true and predicted labels, effectively distinguishing between Arctic ocean and interglacial water, demonstrating good classification performance.

[0032] This invention selects an area near 83.5°N, 20°E where MODIS imagery and the Cryosat-2 satellite trajectory are superimposed as a field comparison. Data with minimal cloud cover and dates close to those of the MODIS imagery and Cryosat-2 satellite trajectory are chosen to minimize the impact of cloud cover and sea ice drift. Figure 6 The black area shown is identified as sea ice, and the red area is identified as interglacial water.

[0033] 2.2 Waveform Backtracking Synthetic Aperture Radar (SAR) is currently the primary satellite altimetry technology applied in the Arctic region, and its accompanying waveform retracking algorithm plays a crucial role in accurately obtaining the true distance between the satellite and the Earth's surface. The Arctic region contains a large amount of sea ice, interglacial water, meltponds, and other complex features, resulting in a diverse and complex waveform representation. Currently, the mainstream waveform retracking methods in the Arctic can be broadly categorized into three types: empirical algorithms (e.g., Peacocket et al. 2004; Gileset et al. 2008; Helmetal et al. 2014), mathematical model methods (e.g., Yiet et al. 2018; Shenet et al. 2019; Landye et al. 2022; Dinardoe et al. 2016), and physical model methods (e.g., Rayet et al. 2015; Laforge et al. 2021). Empirical algorithms are simple to compute and highly efficient. Compared with physical model methods and mathematical model methods, they do not require input of variables such as backscattering coefficient, radar wavelength, and surface roughness, and they do not suffer from high computational complexity or difficulty in fitting.

[0034] This invention uses two empirical algorithms, TFMRA and PPTRA. TFMRA is a widely used empirical algorithm, and many researchers have used it in their studies, demonstrating that it has better accuracy than other retracking methods. TFMRA aims to find the backtracking point when the waveform first rises rapidly, making it suitable for waveforms with strong reflection characteristics and more accurate for interglacial water. PPTRA is an algorithm that combines the features of OCOG (offsetcentreofgravity) and threshold retracker.

[0035] ; ; Where N is the total number of sampling points of the waveform to be calculated. This represents the power at the location of the i-th waveform sampling point. It is the power difference between two adjacent waveform sampling points. It is the power difference between two sampling points that are one sampling point apart. and This method calculates the thresholds at the start and end of the wave peak, respectively. and The value is used to locate the main peak of the waveform. Figure 4 The red portion represents the extracted main peak. For ocean echoes, double or multiple peaks may appear. Unlike the OCOG algorithm or threshold re-tracking method, which are applied to the complete waveform, the PPTRA algorithm only considers finding the "main peak" of the echo, thus providing better accuracy for ocean waveforms. This invention improves upon the original PPTRA algorithm to be suitable for the latest version of Cryptosat-2 SAR data, Baseline-E. In versions prior to Baseline-E, the number of range gates in the SAR mode was halved. Baseline-E data, to improve waveform accuracy, increases the sampling frequency and doubles the length of the range gates. Therefore, this invention also optimizes previous methods. Specifically, for the range of the main peak, this invention selects... The first greater than and the first less than The positions of the first and last 8 bins were used as the start and end points of the main peak to accommodate the longer sampling interval of Cryosat-2SAR mode data. In addition, the main peak position was oversampled by 10 times to improve the accuracy of the retracking point. Figure 5 This is a schematic diagram of the TFMRA and PPTRA algorithms. The red markers indicate the re-tracking points determined by the two algorithms, while the red line segments indicate the main peak intervals identified by the PPTRA algorithm.

[0036] 2.3 Sea Surface Dynamic Topography Survey By using the location of the retracking point, a more accurate sea surface height (SSH) at the satellite footpoint can be calculated, and then the DOT in the Arctic Ocean can be calculated. (1) (2) Where Orbit is the satellite's orbital altitude relative to the WGS84 ellipsoid, extracted from satellite data, and Range is the distance from the satellite to the Earth's surface after retracking calculations. The sea level height is obtained by subtracting the distance between the satellite and the instantaneous sea surface from the satellite's altitude relative to the WGS84 ellipsoid. Yes, tidal and atmospheric geophysical corrections need to be removed from sea level calculations. Calculating sea surface dynamic topography also requires the height of the geoid. Here, we use the GOCO06s geoid.

[0037] For Prandi's PADT dataset, since he calculates absolute dynamic terrain (ADT) instead of traditional sea surface dynamic terrain, additional calculations are required. First, the sea surface height anomaly SLA is calculated according to formula (3) (Cheltonetal.2001). The mean sea level (MSS) needs to be subtracted from the calculated sea surface height. For absolute dynamic terrain ADT, the mean dynamic terrain (MDT) only needs to be added to formula (3): (3) (4) In calculating the absolute dynamic topography (ADT), this invention uses the latest mean sea level (MSS) and mean dynamic topography (MDT) models published by AVISO. These models are improved upon compared to previous models and updated with data from recent years, exhibiting smaller errors at nearshore costal locations. The selection of geographical corrections is shown in Table 2.

[0038] Table 2 Corrections for the Altimetry Data Study

[0039] 2.4 Data Editing The Cryosat-2 satellite has a 30-day subcycle, which can cover the Arctic Ocean within one subcycle, meeting the needs for constructing monthly sea surface dynamic topography data for the Arctic Ocean. However, Cryosat-2 data does not cover all areas of the Arctic Ocean, and there are many sea ice areas. These areas need to be supplemented by interpolation of nearby interglacial water. A 25km EASE2 projection grid is constructed. It is an equal-area projection, which is particularly important in high-latitude regions. First, outliers are filtered out based on a range of ±2. Then, the DOT is calculated, and error values ​​that differ from the three-month average grid by 2.5 times σ are excluded. Finally, the calculated results are smoothed with a Gaussian surface roughness of 100km, and Kriging interpolation is used to supplement these areas without data and sea ice areas. Figure 6 This is a monthly map showing the dynamic ocean topography of the Arctic in March each year from 2011 to 2024. The red areas clearly show the extent of the Beaufort Circulation, which is most pronounced in the Canada Basin.

[0040] 3. Data accuracy assessment This invention will introduce the data accuracy of this study, because the sealevel of the ice-covered area of ​​the Arctic Ocean is obtained by interpolation, so it needs to be compared with some published independent datasets or tide gauge data to verify the accuracy.

[0041] 3.1 Comparison with Arctic Sea Surface Height Data Set In previous studies, Armitage et al. (2016) and Prandie et al. (2021) used the Cryptosat-2 satellite to calculate the dynamic or absolute dynamic topography of the Arctic Ocean and produced CPOM and PADT. These data were then compared using the EASE2 25km grid coordinate system. The results are shown in […]. Figure 7 and Figure 8 . Figure 7 (a) represents the difference in mean sea surface dynamic topography between the CryoSat-2 and CPOM datasets; (b) represents the variance ratio of sea surface dynamic topography. Figure 8 (a) Mean absolute dynamic terrain difference between CryoSat-2 data and PADT dataset; (b) Variance ratio of absolute dynamic terrain.

[0042] The geoid model used in calculating DOT is GOCO06s, which differs from GOCO03s used in CPOM. This results in an overall larger average DOT compared to CPOM. Furthermore, the MSS model used in calculating ADT differs from the DTU15MSS used in PADT, leading to an overall smaller average ADT. Since this study uses a single satellite, Cryosat-2, without cross-calibration with other satellite data, the data consistency in low-latitude regions exhibits significant fluctuations. These regions lack long-term Cryosat-2 data and lack supplementary data fusion, thus being affected by interpolation methods, resulting in larger errors, which are more pronounced than in other regions. However, the comparison of average DOT and variance shows acceptable accuracy in most high-latitude regions.

[0043] 3.2 Comparison with data from tide gauge stations Tide gauge stations are an important means of assessing the accuracy of satellite data, providing precise data for local areas as a reference. However, sea level measurements taken by tide gauge stations are based on local reference planes, which are difficult to reconcile with the reference planes used for satellite altimetry. Therefore, this invention selects sea level anomalies within a 125-kilometer radius of the tide gauge station for comparison. The tide gauge station data is analyzed monthly, and missing data is imputed using linear interpolation. Corrections and RMSd (Root Mean Square Deviation) are compared between the satellite data and the tide gauge station data. The comparison results are shown in Table 3, with RMSd values ​​in meters.

[0044] Table 3 Comparison of each dataset with observations from tide gauge stations

[0045] Figure 9 A comparative chart of SLA time series data for sea level height anomalies from three tide gauge stations: Kigliah, Sterlegova, and Alert (where tide gauge station data is dark gray, CryoSat-2 (CS2) is red, PADT-CS2 is green, CPOM is purple, and there is also purple PADT fusion data). Figure 9In this study, three tide gauge stations with substantial data matching to all four datasets were selected, and SLA time series plots for 2011-2014 were generated. The CS2 dataset showed generally consistent variability with other altimetry products and tide gauge records, particularly in capturing major interannual and seasonal fluctuations. Among these stations, the CryoSat-2 series exhibited amplitude and phase comparable to the PADT-CS2 and CPOM datasets. By comparing the SLA of sea level anomalies at the tide gauge stations, it was found that the CS2 dataset closely matches the long-term trends indicated by tide gauges, validating its accuracy in calculating sea level height.

[0046] Based on the comparison results with tide gauge stations in the table, the data of this invention (CS-2) is generally superior to the CPOM data in terms of correlation R and root mean square deviation (RMSd), exhibiting higher consistency and lower error levels. Compared with PADT-CS2, the data of this invention shows similar accuracy at most tide gauge stations, especially in terms of RMSd, where the difference is small. Furthermore, PADT-CS2 data is based on sea ice fissure (interglacial water) point extraction, resulting in some data gaps in the Arctic region; the data of this invention supplements the data for these locations. In the comparison with PADT-Multimission, although the data of this invention has lower correlation than the fused data, the difference in RMSd is controlled within 0.1 at most stations. This also indicates that the data of this invention has the potential to serve as reference or supplementary data for multi-source satellite fusion, especially in areas where fused data coverage is limited or missing; the single-satellite processing results of this invention can still provide valuable sea surface height information.

[0047] This invention updates the latest Arctic Ocean sea level dataset from the Cryosat-2 satellite from 2011 to 2024. The main processing methods include unsupervised classification and gradient boosting to classify the Arctic Ocean sea level. The TFMRA and PPTRA algorithms are combined to re-track interglacial water and ocean data in the Arctic Ocean to obtain the Arctic Ocean sea level data. The resulting Arctic Ocean sea level anomaly (SLA) and sea surface dynamic topography (DOT) data cover all ocean and ice-covered areas within the latitude range of 60-85°N. The data grid resolution is 25 km, and the time range is one data point per month from 2011 to 2024. The difference between the average DOT and CPOM and PADT is at a reasonable level. In areas where Cryosat-2 data is lacking, the variance ratio of the data from this invention reaches its maximum. The accuracy of the data from this invention is improved compared to CPOM sea level data and comparable to the accuracy of data calculated by PADT using only the CS2 satellite. Ice fills in the missing areas, indicating that the data from this invention can provide the latest data support for Arctic Ocean sea level research and the fusion of satellite data.

[0048] The dynamic topography of the Arctic sea surface reflects the characteristics of ocean circulation and is crucial for monitoring the spatiotemporal changes of polar mesoscale ocean circulation. It can be used to further analyze sea ice transport, freshwater changes, and ice-sea-atmosphere interaction mechanisms, providing constraints for polar climate system modeling. Moreover, the dataset is based on CryoSat-2 and can be fused with satellite data such as Sentinel-3 and ICESat-2 in the future to improve spatiotemporal continuity and provide more complete observations for Arctic Ocean dynamics research. It can be used to study the long-term trends of Arctic Ocean sea surface height and dynamic topography, especially signals related to climate change and accelerated polar warming (arcticamplification).

[0049] Based on the Cryosat-2 satellite, the K-Medoids clustering method is first used to perform initial classification of Arctic Ocean waveforms without any prior knowledge. Then, the gradient boosting algorithm is combined to further classify the clustering results. This unsupervised classification combined with supervised classification method produces a model that is suitable for classifying the large number and complex radar waveforms in the Arctic Ocean and has high classification accuracy. Different retracking methods are used for the two types of Arctic Ocean sea surface waveforms. For interglacial water echoes with rapidly changing power, the Threshold First-Maximum Retracker Algorithm (TFMRA) is used, which can quickly find the retracking point when the waveform power rises rapidly. For open ocean echoes with slow changes at the trailing edge and complex situations with double or multiple peaks, the Primary Peak Threshold Retracker Algorithm (PPTRA) is used, which can extract the main peak of the waveform while ignoring the influence of other peaks. The combination of the two retracking algorithms enables more detailed calculation of Arctic Ocean sea surface height.

[0050] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for monitoring sea surface height in ice-covered sea areas based on satellite altimetry data, characterized in that, The method comprises the following steps: S1, acquiring SAR mode radar echo data of CryoSat-2 satellite in ice-covered sea area, and synchronously acquiring sea ice concentration data and optical remote sensing image data; S2, based on a plurality of characteristic parameters of the SAR mode radar echo data, using an unsupervised clustering method to perform initial classification on the waveforms, and then combining a gradient boosting tree model to optimize the classification result and perform classification processing on the waveforms; S3, according to the waveform classification result, using different empirical re-tracking algorithms for distance correction on waveforms of different types; S4, based on the distance data after re-tracking, combining satellite orbit height, geophysical correction terms and geoid model to calculate sea surface height data; S5, projecting the sea surface height data to an equal-area grid, performing spatial interpolation and smoothing processing on the missing area to generate a monthly sea surface height data set covering the ice-covered sea area; S6, comparing and verifying the generated sea surface height data with the tide station data and the existing data set, and outputting the final sea surface height monitoring result.

2. The method for monitoring sea surface height of ice-covered sea area based on satellite altimetry data according to claim 1, characterized in that, In step S2, the K-medoids clustering algorithm is used to perform unsupervised initial classification on the radar echo waveforms, and based on the clustering result, the waveforms are merged into four categories of marine waveforms, inter-ice water waveforms, sea ice waveforms and undefined waveforms; The merged classification result is used as training data and input into the gradient boosting tree model for supervised classification.

3. The method for monitoring sea surface height of ice-covered sea area based on satellite altimetry data according to claim 1, characterized in that, In step S3, the maximum threshold re-tracking algorithm is used for inter-ice water waveforms to locate the re-tracking point of the first rapid rise of the waveforms; and the main peak threshold re-tracking algorithm is used for marine waveforms to locate the main peak position of the waveforms.

4. The method for monitoring sea surface height of ice-covered sea area based on satellite altimetry data according to claim 3, characterized in that, The main peak threshold re-tracking algorithm locates the main peak position of the waveform by calculating the threshold at the beginning and end of the calculated wave peak and ​ calculating a power difference between two samples separated by one sample point : ； Computing a threshold at the beginning of a wave crest : ； calculating a power difference between two adjacent waveform sample points : ； Computing a threshold at the end of a wave peak : ； Wherein, N is the total sampling points of the waveform to be calculated, represents the power at the i-th waveform sampling point position, represents the power at the i+1-th waveform sampling point position.

5. The method for monitoring sea surface height of ice-covered sea area based on satellite altimetry data according to claim 3, characterized in that, The sea surface height data includes sea surface height SSH, dynamic topography of sea surface DOT and absolute dynamic topography ADT; ； ； ； where Orbit is the orbit height of the satellite relative to the WGS84 ellipsoid; Range is the distance from the satellite to the surface of the earth object after re-tracking calculation; is the tidal and atmospheric geophysical correction; is the height of the geoid; SLA is the sea level anomaly; and MDT is the mean dynamic topography.

6. The method for monitoring sea surface height of ice-covered sea area based on satellite altimetry data according to claim 5, characterized in that, The tide and geophysical correction includes orbit correction, ionospheric correction, wet troposphere correction, dry troposphere correction, dynamic atmosphere correction, solid tide, ocean tide and polar tide correction.

7. The method for monitoring sea surface height of ice-covered sea area based on satellite altimetry data according to claim 5, characterized in that, The sea level anomaly SLA is: ； Wherein, MSS is the mean sea level.

8. The method for monitoring sea surface height of ice-covered sea area based on satellite altimetry data according to claim 1, characterized in that, In step S5, the equal-area grid is a 25km resolution EASE2 projection grid; the spatial interpolation is a Kriging interpolation method; and the smoothing processing is a 100km scale Gaussian smoothing.

9. The method for monitoring sea surface height of ice-covered sea area based on satellite altimetry data according to claim 1, characterized in that, In step S6, the data verification includes: comparing the generated sea surface height data with the tide station data provided by PSMSL, CPOM data set and / or PADT data set, and evaluating the data accuracy by calculating the correlation coefficient and root mean square deviation.

Images