A GNSS-R-based sea ice density inversion model training method and device

Abstract

The application provides a GNSS-R-based sea ice density inversion model training method and device, comprising: acquiring training data according to a preset time interval, wherein the training data comprises GNSS-R data and real sea ice density that is spatio-temporally aligned with the GNSS-R data; determining the ground surface type of a corresponding target area based on the GNSS-R data, wherein the ground surface type is water surface or ice surface; inputting the GNSS-R data of the ice surface into a sea ice density inversion model to obtain predicted sea ice density of the corresponding target area through sea ice density inversion; determining a loss value between the real sea ice density and the predicted sea ice density, and iteratively adjusting model parameters of the sea ice density inversion model according to the loss value to obtain an updated sea ice density inversion model. The sea ice density inversion model focuses more on feature learning of ice surface GNSS-R data, and the accuracy of sea ice density inversion is improved.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a training method and apparatus for a sea ice concentration inversion model based on GNSS-R. Background Technology

[0002] Sea ice concentration refers to the percentage of the area covered by sea ice in a specific sea area. It is not only an important input parameter for climate models, but also of great significance to the safety of Arctic shipping routes.

[0003] In existing technologies, GNSS-R (Global Navigation Satellite System Reflectometry) technology can be used to invert sea ice concentration. After the navigation satellite transmits an L-band signal to the Earth's surface, the reflected signal of the L-band signal from the Earth's surface is received. Since seawater and sea ice have different reflection characteristics, the sea ice concentration can be inverted by analyzing the characteristic differences of these reflected signals through an inversion model.

[0004] However, the movement and changes of sea ice are very complex, especially in the Antarctic region where the state of sea ice is constantly changing. Current inversion models cannot accurately capture these complex changes, resulting in significant uncertainty and large errors in the inversion results of sea ice concentration. Summary of the Invention

[0005] To address the aforementioned technical problems, this application presents a training method and apparatus for a GNSS-R-based sea ice concentration inversion model, which solves the problem that the inversion model cannot accurately capture the complex changes in sea ice, resulting in significant uncertainty and large errors in the sea ice concentration inversion results.

[0006] Firstly, this application presents a method for training a GNSS-R-based sea ice concentration inversion model, including:

[0007] Training data is acquired at preset time intervals, including Global Navigation Satellite System Reflection GNSS-R data and real sea ice concentrations spatiotemporally aligned with the GNSS-R data.

[0008] Based on the GNSS-R data, the surface category of the corresponding target area is determined, which is either water or ice.

[0009] The GNSS-R data with the surface category of ice is input into the sea ice concentration inversion model to perform sea ice concentration inversion and obtain the predicted sea ice concentration of the corresponding target area.

[0010] The loss value between the actual sea ice concentration and the predicted sea ice concentration is determined, and the model parameters of the sea ice concentration inversion model are iteratively adjusted according to the loss value to obtain the updated sea ice concentration inversion model.

[0011] Optionally, acquiring training data at preset time intervals includes:

[0012] GNSS-R data and actual sea ice concentration are acquired at preset time intervals;

[0013] Based on latitude and longitude coordinates and time information, bilinear interpolation is used to perform spatiotemporal alignment of the GNSS-R data and the actual sea ice concentration to construct training data.

[0014] Optionally, acquiring GNSS-R data and actual sea ice concentration at preset time intervals includes:

[0015] Acquire raw data, which includes sample GNSS-R data and sample real sea ice concentration;

[0016] A sliding time window is used to collect GNSS-R data and real sea ice concentration from the raw data. The data collection duration of each sliding time window is the preset time interval.

[0017] Optionally, determining the surface category of the corresponding target area based on the GNSS-R data includes:

[0018] Obtain historical sea ice concentration, and based on the historical sea ice concentration, calculate the average sea ice concentration of the target area corresponding to the GNSS-R data within a preset time period;

[0019] If the average sea ice concentration is below a first threshold, the surface category of the target area is determined to be water surface;

[0020] If the average sea ice concentration is higher than the second threshold, the surface category of the target area is determined to be ice surface;

[0021] If the average sea ice concentration is higher than or equal to the first threshold and lower than or equal to the second threshold, the surface category of the corresponding target area is determined based on the GNSS-R data.

[0022] Optionally, determining the surface category of the corresponding target area based on the GNSS-R data includes:

[0023] Extract the number of valid pixels from the GNSS-R data;

[0024] If the number of effective pixels is less than a preset threshold, the surface type of the target area is determined to be ice.

[0025] If the number of effective pixels is greater than or equal to the preset threshold, the surface category of the target area is determined to be water.

[0026] Optionally, the GNSS-R data includes a time-delay Doppler map, where each pixel in the time-delay Doppler map has a corresponding power; extracting the effective pixel count from the GNSS-R data includes:

[0027] Determine the power peak value in the time-delay Doppler plot;

[0028] The number of pixels whose power exceeds a preset percentage of the power peak value is determined as the effective pixel count.

[0029] Optionally, the GNSS-R data includes satellite numbers. The step of inputting the GNSS-R data with a surface category of ice into a sea ice concentration inversion model to perform sea ice concentration inversion and obtain the predicted sea ice concentration for the corresponding target area includes:

[0030] Based on the satellite number, the navigation system corresponding to each GNSS-R data with the surface category of ice is determined, resulting in multiple groups, each group including multiple GNSS-R data;

[0031] For each group, feature analysis is performed on each GNSS-R data within the group to obtain at least one corresponding feature parameter;

[0032] The feature parameters are input into the sea ice concentration inversion model corresponding to each group to perform sea ice concentration inversion and obtain the predicted sea ice concentration of the corresponding target area.

[0033] Optionally, determining the loss value between the actual sea ice concentration and the predicted sea ice concentration includes:

[0034] Determine the mean square error, mean absolute error, and correlation coefficient between the actual sea ice concentration and the predicted sea ice concentration;

[0035] The loss value is determined based on the mean square error, the mean absolute error, and the correlation coefficient.

[0036] Optionally, after obtaining the updated sea ice concentration inversion model, the method further includes:

[0037] The GNSS-R data whose corresponding real sea ice concentration values ​​are within the preset range are used as verification data;

[0038] The verification data is input into the updated sea ice concentration inversion model to perform sea ice concentration inversion and obtain the updated sea ice concentration of the corresponding target area.

[0039] A top-down view of the sea ice distribution in the target area is generated based on the updated sea ice concentration. This top-down view is used to verify the updated sea ice concentration inversion model.

[0040] Secondly, this application discloses a training device for a GNSS-R-based sea ice concentration inversion model, comprising:

[0041] The acquisition module is used to acquire training data at preset time intervals. The training data includes Global Navigation Satellite System Reflection (GNSS-R) data and real sea ice concentrations that are spatiotemporally aligned with the GNSS-R data.

[0042] The classification module is used to determine the surface category of the corresponding target area based on the GNSS-R data, wherein the surface category is water or ice.

[0043] The prediction module is used to input the GNSS-R data with the surface category of ice into the sea ice concentration inversion model to perform sea ice concentration inversion and obtain the predicted sea ice concentration of the corresponding target area.

[0044] An adjustment module is used to determine the loss value between the actual sea ice concentration and the predicted sea ice concentration, and to iteratively adjust the model parameters of the sea ice concentration inversion model based on the loss value to obtain the updated sea ice concentration inversion model.

[0045] Optionally, the acquisition module is specifically used for:

[0046] GNSS-R data and actual sea ice concentration are acquired at preset time intervals;

[0047] Based on latitude and longitude coordinates and time information, bilinear interpolation is used to perform spatiotemporal alignment of the GNSS-R data and the actual sea ice concentration to construct training data.

[0048] Optionally, the acquisition module is specifically used for:

[0049] Acquire raw data, which includes sample GNSS-R data and sample real sea ice concentration;

[0050] A sliding time window is used to collect GNSS-R data and real sea ice concentration from the raw data. The data collection duration of each sliding time window is the preset time interval.

[0051] Optionally, the classification module is specifically used for:

[0052] Obtain historical sea ice concentration, and based on the historical sea ice concentration, calculate the average sea ice concentration of the target area corresponding to the GNSS-R data within a preset time period;

[0053] If the average sea ice concentration is below a first threshold, the surface category of the target area is determined to be water surface;

[0054] If the average sea ice concentration is higher than the second threshold, the surface category of the target area is determined to be ice surface;

[0055] If the average sea ice concentration is higher than or equal to the first threshold and lower than or equal to the second threshold, the surface category of the corresponding target area is determined based on the GNSS-R data.

[0056] Optionally, the classification module is specifically used for:

[0057] Extract the number of valid pixels from the GNSS-R data;

[0058] If the number of effective pixels is less than a preset threshold, the surface type of the target area is determined to be ice.

[0059] If the number of effective pixels is greater than or equal to the preset threshold, the surface category of the target area is determined to be water.

[0060] Optionally, the GNSS-R data includes a time-delay Doppler map, where each pixel in the time-delay Doppler map has a corresponding power; the classification module is specifically used for:

[0061] Determine the power peak value in the time-delay Doppler plot;

[0062] The number of pixels whose power exceeds a preset percentage of the power peak value is determined as the effective pixel count.

[0063] Optionally, the GNSS-R data includes satellite identification numbers, and the prediction module is specifically used for:

[0064] Based on the satellite number, the navigation system corresponding to each GNSS-R data with the surface category of ice is determined, resulting in multiple groups, each group including multiple GNSS-R data;

[0065] For each group, feature analysis is performed on each GNSS-R data within the group to obtain at least one corresponding feature parameter;

[0066] The feature parameters are input into the sea ice concentration inversion model corresponding to each group to perform sea ice concentration inversion and obtain the predicted sea ice concentration of the corresponding target area.

[0067] Optionally, the adjustment module is specifically used for:

[0068] Determine the mean square error, mean absolute error, and correlation coefficient between the actual sea ice concentration and the predicted sea ice concentration;

[0069] The loss value is determined based on the mean square error, the mean absolute error, and the correlation coefficient.

[0070] Optionally, the device further includes a verification module for:

[0071] The GNSS-R data whose corresponding real sea ice concentration values ​​are within the preset range are used as verification data;

[0072] The verification data is input into the updated sea ice concentration inversion model to perform sea ice concentration inversion and obtain the updated sea ice concentration of the corresponding target area.

[0073] A top-down view of the sea ice distribution in the target area is generated based on the updated sea ice concentration. This top-down view is used to verify the updated sea ice concentration inversion model.

[0074] Thirdly, this application discloses an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of a GNSS-R-based sea ice concentration inversion model training method as described in any of the preceding claims.

[0075] Fourthly, this application discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the training method for a GNSS-R-based sea ice concentration inversion model as described in any of the preceding claims.

[0076] Compared with the prior art, this application has the following advantages:

[0077] In this application, training data containing GNSS-R data and spatiotemporal alignment of real sea ice concentration is acquired at preset time intervals to capture the dynamic changes of sea ice over time. Then, based on the GNSS-R data, the surface category of the target area is determined to be water or ice. Only the GNSS-R data of the ice surface is input into the sea ice concentration inversion model for inversion to obtain the predicted sea ice concentration. Furthermore, the model parameters of the sea ice concentration inversion model are iteratively adjusted by the loss value between the real sea ice concentration and the predicted sea ice concentration, so that the sea ice concentration inversion model focuses more on the feature learning of the ice surface GNSS-R data, reduces the interference of water surface GNSS-R data, thereby improving the ability to capture complex changes in sea ice, reducing the uncertainty and error of the inversion results, and improving the accuracy of sea ice concentration inversion. Attached Figure Description

[0078] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0079] Figure 1 This is a flowchart illustrating the steps of a GNSS-R-based sea ice concentration inversion model training method according to this application.

[0080] Figure 2 This is a distribution diagram showing the correspondence between reflectance and sea ice concentration in one embodiment of this application;

[0081] Figure 3 This is a comparison diagram of the time delay Doppler images corresponding to GPS, BDS and GALILEO respectively in one embodiment of this application;

[0082] Figure 4 This is a logical schematic diagram of a GNSS-R-based sea ice concentration inversion model training method in one embodiment of this application;

[0083] Figure 5 This is a comparison chart of the mean squared error and correlation coefficient calculated in one embodiment of this application;

[0084] Figure 6 This is a structural block diagram of a GNSS-R-based sea ice concentration inversion model training device according to this application;

[0085] Figure 7 This is a structural block diagram of an electronic device according to this application. Detailed Implementation

[0086] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0087] Sea ice concentration refers to the percentage of the area covered by sea ice in a specific sea area. Among related technologies, GNSS-R (Global Navigation Satellite System Reflectometry) technology can be used to invert sea ice concentration. After the navigation satellite transmits L-band signals to the Earth's surface, the reflected signals of the L-band signals from the Earth's surface are received. Since seawater and sea ice have different reflection characteristics, the sea ice concentration can be inverted by analyzing the characteristic differences of these reflected signals through an inversion model.

[0088] However, the movement and changes of sea ice are highly complex, especially in the Antarctic region where sea ice conditions are constantly changing. Current inversion models cannot accurately capture these complex changes, leading to significant uncertainty and large errors in the sea ice concentration inversion results. Therefore, this application proposes a GNSS-R-based sea ice concentration inversion model training method to address these issues.

[0089] The following will describe in detail a training method for a sea ice concentration inversion model based on GNSS-R provided by the present invention through specific embodiments.

[0090] Reference Figure 1 The diagram illustrates a step-by-step flowchart of a GNSS-R-based sea ice concentration inversion model training method, which specifically includes the following steps:

[0091] S11: Acquire training data at preset time intervals. The training data includes Global Navigation Satellite System Reflection (GNSS-R) data and real sea ice concentrations that are spatiotemporally aligned with the GNSS-R data.

[0092] like Figure 2 As shown, by analyzing the distribution map of the correspondence between a certain characteristic parameter (reflectivity) and sea ice concentration in summer (e.g., August 2023) and winter (e.g., March 2024), it can be clearly observed that there are huge differences between the two in spatial structure and numerical distribution. This seasonal variation poses a challenge to the generalization ability of the sea ice concentration inversion model, causing a GNSS-R-based sea ice concentration inversion model trained on training data of a single season to often be difficult to adapt to the characteristic distribution of other seasons.

[0093] Therefore, in this step, training data is continuously collected at preset time intervals, aiming to capture the dynamic changes of sea ice over time through time-series-based sample coverage. The training data consists of two parts:

[0094] Global Navigation Satellite System Reflection (GNSS-R) data mainly comes from the observation of surface reflection signals by signal receiving equipment. The signal receiving equipment can be the FY-3E GNOS-II (Global Navigation Satellite Occultation Sounder – II). GNOS-II is a remote sensing instrument carried on the Chinese polar-orbiting meteorological satellite FY-3E. It supports the collection of surface reflection signals of the bands emitted by navigation satellites of various systems such as GPS (Global Positioning System), BDS (BeiDou Navigation Satellite System), and Galileo (Galileo Satellite Navigation System).

[0095] Real sea ice concentration data that is spatiotemporally aligned with GNSS-R data, where the real sea ice concentration data typically comes from the Level 3 sea ice concentration product provided by OSI SAF (Ocean and Sea Ice Satellite Application Facility).

[0096] In the training data, the GNSS-R data and the actual sea ice concentration have been spatiotemporally aligned. In other words, each set of GNSS-R data can be precisely matched with the actual sea ice concentration in the same area at the same time, thus providing a data foundation that is both timely and spatially consistent for the training of the subsequent sea ice concentration inversion model.

[0097] S12: Based on GNSS-R data, determine the surface category of the corresponding target area, which is either water or ice.

[0098] In this step, the significant differences in the reflection characteristics of seawater and sea ice to L-band signals can be utilized to achieve land surface classification through in-depth analysis of GNSS-R data.

[0099] Specifically, sea ice surfaces are relatively smooth, resulting in highly concentrated reflected signal energy, high peak power, and a narrow spatial range; while seawater surfaces are typically rough, causing reflected signal energy to be spatially dispersed, resulting in lower peak power and a more pronounced spatial spread. Furthermore, the equivalent reflectivity of seawater is generally lower than that of sea ice, while the signal-to-noise ratio may exhibit opposite trends due to differences in surface roughness, and so on.

[0100] Based on the aforementioned physical differences, various algorithms can be used to classify land surface categories based on GNSS-R data, thereby filtering out GNSS-R data related to the ice surface. This provides targeted input for the subsequent training of the sea ice concentration inversion model, improving the efficiency and accuracy of the sea ice concentration inversion model training.

[0101] S13: Input GNSS-R data with the surface category of ice into the sea ice concentration inversion model to perform sea ice concentration inversion and obtain the predicted sea ice concentration of the corresponding target area.

[0102] In this step, GNSS-R data with the surface category of ice can be input into the sea ice concentration inversion model to be trained. The sea ice concentration inversion model will make a preliminary prediction of the sea ice concentration in the target area and output the corresponding predicted sea ice concentration (range 0%-100%).

[0103] Among them, the sea ice concentration inversion model can be an SVR (Support Vector Regression) model, which uses an RBF (Radial Basis Function kernel) kernel to establish a nonlinear mapping between GNSS-R data with the surface category of ice and the predicted sea ice concentration. Then, after inputting GNSS-R data with the surface category of ice into the sea ice concentration inversion model, the sea ice concentration inversion model can output the predicted sea ice concentration of the target area.

[0104] It is understandable that the core of sea ice concentration inversion is to infer the ice coverage ratio from the features of GNSS-R data. However, the reflectivity of water and ice surfaces differs significantly. If GNSS-R data of the water surface is included in the training, the sea ice concentration inversion model may incorrectly learn the correlation between water surface features and sea ice concentration, leading to a decrease in the accuracy of ice area inversion.

[0105] Therefore, by focusing on ice surface GNSS-R data, the sea ice concentration inversion model can learn the intrinsic relationship between ice surface GNSS-R data and sea ice concentration, avoiding interference from water surface GNSS-R data on the sea ice concentration inversion model's learning of ice surface characteristics, thereby enhancing the sea ice concentration inversion model's ability to capture ice surface reflection patterns.

[0106] S14: Determine the loss value between the actual sea ice concentration and the predicted sea ice concentration, and iteratively adjust the model parameters of the sea ice concentration inversion model based on the loss value to obtain the updated sea ice concentration inversion model.

[0107] In this step, the model parameters can be iteratively adjusted by calculating the loss value between the actual sea ice concentration and the predicted sea ice concentration, and finally the optimized sea ice concentration inversion model can be obtained.

[0108] Specifically, the calculation of the loss value aims to quantify the deviation between the actual sea ice concentration and the predicted sea ice concentration. If the difference between the actual sea ice concentration and the predicted sea ice concentration is large, a large loss value will be generated; conversely, if the difference between the actual sea ice concentration and the predicted sea ice concentration is small, the loss value will be low.

[0109] Based on the obtained loss value, the sea ice concentration inversion model can continuously fine-tune its own model parameters (such as kernel function parameters and penalty coefficients in the support vector regression model) through the backpropagation mechanism, gradually reduce the prediction bias, and improve the inversion accuracy of the sea ice concentration inversion model in a targeted manner, eventually converging into an updated sea ice concentration inversion model that is more suitable for complex sea ice conditions.

[0110] In one implementation, training data is acquired at preset time intervals, including:

[0111] GNSS-R data and actual sea ice concentration are acquired at preset time intervals;

[0112] Based on latitude and longitude coordinates and time information, bilinear interpolation is used to spatiotemporally align GNSS-R data and real sea ice concentration to construct training data.

[0113] In this implementation, GNSS-R data and actual sea ice concentration are first collected at preset time intervals. Each set of GNSS-R data corresponds to a specific "reflection point" on the Earth's surface. The latitude and longitude coordinates of the reflection point can be calculated using satellite orbit parameters, signal propagation paths, etc., to accurately identify the spatial location of the signal source. Correspondingly, actual sea ice concentration is usually presented in a gridded form, with each grid cell corresponding to fixed latitude and longitude coordinates (e.g., 10km × 10km), clearly defining the geographical area covered by the latitude and longitude coordinates. The time information of GNSS-R data and actual sea ice concentration is directly linked to their reception or update time, reflecting the specific time when the observation occurred. The time information can be based on UTC (Coordinated Universal Time) or other time bases, without specific limitations.

[0114] Since there may be differences in spatial resolution or temporal granularity between the acquisition of GNSS-R data and real sea ice concentration data, it is necessary to construct training samples through spatiotemporal alignment processing.

[0115] Specifically, a bilinear interpolation method can be used to accurately align GNSS-R data and actual sea ice concentration in space and time by weighted averaging of adjacent data points. Spatially, interpolation maps the grid information of actual sea ice concentration to the specific location of the GNSS-R reflection point. Temporally, interpolation adjustments are made according to the UTC time to ensure that the two correspond within the same spatiotemporal range. After bilinear interpolation, GNSS-R data and actual sea ice concentration achieve precise matching in both time and space.

[0116] Furthermore, quality control and screening can be performed on GNSS-R data and real sea ice concentration, including outlier removal, data interpolation, noise reduction, normalization, etc., without any specific limitations.

[0117] This process allows for the construction of training data that meets the training requirements of the sea ice concentration inversion model, providing precisely matched input data for subsequent training of the sea ice concentration inversion model.

[0118] In one implementation, GNSS-R data and actual sea ice concentration are acquired at preset time intervals, including:

[0119] Obtain raw data, including sample GNSS-R data and sample real sea ice concentration;

[0120] A sliding time window is used to collect GNSS-R data and real sea ice concentration from the raw data. The data collection time for each sliding time window is a preset time interval.

[0121] In this implementation, the training data acquisition process relies on a sliding time window mechanism. By systematically extracting the raw data, it ensures that the training data can reflect the changing characteristics of sea ice at different time scales.

[0122] First, raw data containing sample GNSS-R data and sample true sea ice concentration needs to be collected in advance. The sample GNSS-R data and sample true sea ice concentration usually include multiple sets of data over a relatively long period of time.

[0123] In order to acquire training data at preset time intervals, a sliding time window is used to extract raw data. The duration of a single data acquisition in the sliding time window is set to a preset time interval (e.g., every half month is a window). By moving the sliding time window on the time axis, GNSS-R data and the corresponding real sea ice concentration can be extracted from the raw data in each sliding time window.

[0124] This sliding mechanism can continuously and evenly cover the entire time series of the original data, ensuring the continuity of the training data in terms of time distribution, and ensuring the consistency of time granularity between each batch of training data through a fixed window duration. This provides support for the sea ice concentration inversion model to capture the short-term dynamic changes and long-term trends of sea ice (such as ice condition evolution during seasonal changes).

[0125] In one implementation, based on GNSS-R data, the surface category of the corresponding target area is determined, including:

[0126] Obtain historical sea ice concentration, and based on historical sea ice concentration, calculate the average sea ice concentration of the target area corresponding to the GNSS-R data within a preset time period;

[0127] When the average sea ice concentration is below the first threshold, the surface category of the target area is determined to be water surface;

[0128] When the average sea ice concentration is higher than the second threshold, the surface category of the target area is determined to be ice surface;

[0129] When the average sea ice concentration is higher than or equal to the first threshold and lower than or equal to the second threshold, the surface category of the corresponding target area is determined based on GNSS-R data.

[0130] In this implementation method, historical sea ice concentration can be obtained first, and based on historical sea ice concentration and GNSS-R data, the surface category of the corresponding target area can be determined, clarifying whether the target area belongs to "water surface" or "ice surface".

[0131] Historical sea ice concentration refers to information such as the spatial distribution, coverage, and changing trends of sea ice over a past period. For example, ERA5 (Fifth generation of the ECMWF Reanalysis for the global climate) sea ice coverage data can be used as historical sea ice concentration. ERA5 sea ice coverage data is generated by fusing multi-source data such as satellite observations and buoy monitoring, combined with numerical model simulations and data assimilation techniques. It has the characteristics of wide global coverage, high data accuracy, and complete time series.

[0132] Specifically, based on historical sea ice concentration, the average sea ice concentration of the target area corresponding to GNSS-R data within a preset time period can be calculated. The preset time period needs to be determined in conjunction with the periodicity of sea ice changes. For example, for sea areas with obvious seasonality, the historical sea ice concentration of the same period in the past 10 years (such as December to March of the following year) can be selected, and the average sea ice concentration of the target area during this period can be calculated, i.e., the average sea ice concentration.

[0133] In this way, by aggregating the time dimension of historical sea ice concentration, the impact of short-term fluctuations can be weakened, and the average sea ice concentration is closer to the long-term stable state of the surface of the target area than the historical sea ice concentration at a single moment.

[0134] Subsequently, the average sea ice concentration can be preliminarily determined based on the preset first threshold and second threshold (where the first threshold is lower than the second threshold): when the average sea ice concentration is lower than the first threshold, it indicates that the target area has been characterized by open water for a long period of time, so the target area is directly determined to be a water surface; when the average sea ice concentration is higher than the second threshold, it indicates that the target area has been stably covered by sea ice for a long period of time, so the target area is determined to be an ice surface.

[0135] For example, assuming the first threshold is 10% and the second threshold is 90%, if the average sea ice concentration in January of the past 10 years in a certain area is 95%, then the area is usually covered by ice in January. If the average sea ice concentration is only 5%, then the area is usually covered by water in January.

[0136] For areas where the average sea ice concentration is between the first and second thresholds, it can be identified as a water-ice transition zone. Since the surface conditions in the water-ice transition zone are more affected by short-term changes (such as seasonal melting or freezing processes), the average sea ice concentration alone cannot accurately reflect its real-time attributes. Therefore, the surface category of the corresponding target area can be further determined based on GNSS-R data.

[0137] This time-averaged classification method makes the determination of surface categories more objective and consistent, while making full use of the long-term patterns contained in historical sea ice concentrations, providing a reliable basis for the targeted processing of ice surface GNSS-R data by subsequent sea ice concentration inversion models.

[0138] In one implementation, in step S12, the surface category of the corresponding target area is determined based on GNSS-R data, including:

[0139] Extract the number of valid pixels from GNSS-R data;

[0140] If the number of valid pixels is less than the preset threshold, the surface type of the target area is determined to be ice.

[0141] If the number of valid pixels is greater than or equal to the preset threshold, the surface category of the target area is determined to be water.

[0142] In this implementation, the effective pixel count N extracted from GNSS-R data is introduced as an auxiliary criterion for land surface category classification. The effective pixel count refers to the number of pixels in GNSS-R data that can reliably and accurately reflect the actual state of the land surface. It reflects the scattering characteristics of the land surface and is thus used to distinguish whether the ice-water transition zone is a water surface or an ice surface.

[0143] Specifically, based on the correlation between the number of effective pixels and the surface scattering characteristics, the surface category can be further determined by using a preset threshold:

[0144] Generally, the scattered signal of ice is more concentrated, corresponding to fewer effective pixels; the scattered signal of water is more dispersed, with more effective pixels. Therefore, by comparing the number of effective pixels with a preset threshold, if the number of effective pixels is less than the preset threshold, the target area is classified as ice, and vice versa.

[0145] In this way, the efficiency of the judgment is ensured by simple and direct numerical comparison, which is especially suitable for auxiliary classification when the average sea ice concentration is in the transition range, and effectively improves the accuracy of classification in the ice-water transition zone.

[0146] In one implementation, the GNSS-R data includes a time-delay Doppler map, where each pixel in the time-delay Doppler map has a corresponding power; extracting the effective pixel count from the GNSS-R data includes:

[0147] Determine the power peak in the time-delay Doppler plot;

[0148] The number of pixels whose power exceeds a preset percentage of the peak power is determined as the number of valid pixels.

[0149] In this implementation, the extraction of effective pixel count and the determination of land surface category depend on the power distribution characteristics of the DDM (Delay-Doppler Map), such as... Figure 3 The diagram shows examples of DDM distribution maps for different navigation systems, corresponding to GPS, BDS, and GALILEO from left to right. It can be seen that there are significant differences in the shape, area, and center location of the high-power region (scattering area) in the DDM of different navigation systems. Specifically, the DDM shape of the Galileo satellite is more concentrated, while the GPS DDM exhibits a larger scattering area. Therefore, by quantifying the concentration of GNSS-R data scattering, accurate differentiation of target area attributes can be achieved, thereby improving the adaptability and generalization ability of the sea ice concentration inversion model to different satellite systems.

[0150] Specifically, the power information is first extracted from the DDM: the power values ​​of all pixels in the image are traversed, the maximum value is determined as the power peak, and then the total number of pixels whose power exceeds the preset proportion (such as 30%) of the power peak is counted. This number is the effective number of pixels.

[0151] In this way, the physical characteristics of DDM data are utilized to provide an objective and reproducible basis for land surface classification.

[0152] In one implementation, the GNSS-R data includes satellite identification numbers. GNSS-R data with an ice surface category is input into a sea ice concentration inversion model to perform sea ice concentration inversion, obtaining the predicted sea ice concentration for the corresponding target area, including:

[0153] Based on satellite identification, the navigation system corresponding to each GNSS-R data with an ice surface category is determined, resulting in multiple groups, each group containing multiple GNSS-R data.

[0154] For each group, feature analysis is performed on each GNSS-R data within the group to obtain at least one corresponding feature parameter;

[0155] The feature parameters are input into the sea ice concentration inversion model corresponding to each group to perform sea ice concentration inversion and obtain the predicted sea ice concentration of the corresponding target area.

[0156] In this implementation, the sea ice concentration inversion process fully considers the differences in signal characteristics of different navigation systems, and improves prediction accuracy through group modeling.

[0157] Specifically, firstly, for GNSS-R data with an ice surface as the land surface category, the GNSS-R data can be divided into navigation systems based on the satellite numbers it contains: the satellite number is directly associated with its navigation system. Based on this, all GNSS-R data are divided into multiple independent groups, and the GNSS-R data in each group comes from the same navigation system, ensuring that the GNSS-R data in the group have consistent signal transmission and reception characteristics.

[0158] For each group, feature analysis and extraction are required for the GNSS-R data. Feature parameters reflecting the surface scattering characteristics are extracted from the DDM, including but not limited to power ratio (PP), DDM average scattering energy (DDMA), equivalent reflectivity (Γ), signal-to-noise ratio (SNR), kurtosis (K), and skewness (S). These feature parameters can quantify the differences in reflection after different navigation system signals interact with sea ice and water surface.

[0159] Subsequently, a corresponding sea ice concentration inversion model is matched for each group (i.e., a dedicated model is trained for different navigation systems), and the extracted feature parameters are input into the corresponding sea ice concentration inversion model for inversion calculation. Since there are inherent differences in the signal frequency, coverage, and scattering response of different navigation systems, the dedicated model can specifically fit the mapping relationship between the features of each navigation system and sea ice concentration, and finally output the predicted sea ice concentration value for each target area.

[0160] This method of modeling different navigation systems effectively avoids systematic errors caused by the mixing of signals from different systems, enabling the model to more accurately capture the unique patterns of each navigation system signal in sea ice inversion, thereby improving the overall reliability of the prediction.

[0161] In one implementation, determining the loss value between the actual sea ice concentration and the predicted sea ice concentration includes:

[0162] Determine the mean square error, mean absolute error, and correlation coefficient between the actual sea ice concentration and the predicted sea ice concentration;

[0163] The loss value is determined based on the mean squared error, mean absolute error, and correlation coefficient.

[0164] In this implementation, the loss value is determined by comprehensively considering multi-dimensional error indicators, fully quantifying the deviation between the actual sea ice concentration and the predicted sea ice concentration, and providing a more accurate direction for optimizing the sea ice concentration inversion model.

[0165] Specifically, three types of error indicators can be calculated: First, the root mean squared error (RMSE), which is obtained by calculating the root of the average of the squares of the differences between the actual and predicted sea ice concentrations, and can intuitively reflect the magnitude of the average error; second, the mean absolute error (MAE), which is calculated by calculating the average of the absolute differences between the actual and predicted sea ice concentrations, and reflects the average level of the overall error, and is less affected by extreme values; and third, the correlation coefficient (R), which is used to assess the degree of linear correlation between the actual and predicted sea ice concentrations. The closer the correlation coefficient is to 1, the more consistent the trends of the two are.

[0166] After obtaining the above three indicators, they can be fused using preset weights or a fusion algorithm to obtain the loss value between the actual sea ice concentration and the predicted sea ice concentration. For example, corresponding weights can be assigned according to the degree of influence of different indicators on the performance of the sea ice concentration inversion model, or the indicators can be mapped to the same order of magnitude through normalization and then summed, etc., without any specific limitations.

[0167] This multi-indicator fusion loss value calculation method avoids the one-sidedness that may exist with a single indicator, and can comprehensively reflect the performance of the prediction results in terms of error magnitude, distribution pattern and correlation. It provides a more comprehensive basis for the iterative adjustment of model parameters and promotes the optimization of the sea ice concentration inversion model towards a direction that is more in line with the real surface state.

[0168] In one implementation, after obtaining the updated sea ice concentration inversion model, the following steps are also included:

[0169] GNSS-R data with corresponding actual sea ice concentration values ​​within a preset range will be used as verification data.

[0170] The verification data is input into the updated sea ice concentration inversion model to perform sea ice concentration inversion and obtain the updated sea ice concentration of the corresponding target area.

[0171] A top-down view of the sea ice distribution in the target area is generated based on the updated sea ice concentration. This top-down view is used to validate the updated sea ice concentration inversion model.

[0172] In this implementation, the updated sea ice concentration inversion model needs to undergo further verification, and its ability to characterize sea ice distribution is intuitively evaluated through spatial visualization.

[0173] Specifically, firstly, GNSS-R data with corresponding actual sea ice concentration values ​​within a preset range can be selected as validation data. This preset range can be set according to actual application needs and the key areas for sea ice monitoring. For example, focusing on the 15%-85% ice-water transition zone, because the sea ice conditions in this area are complex and the inversion is difficult, requiring higher accuracy from the sea ice concentration inversion model. Selecting GNSS-R data within this range as validation data can more effectively test the sea ice concentration inversion model's ability to invert complex sea ice conditions.

[0174] Subsequently, these validation data are input into the updated sea ice concentration inversion model to perform sea ice concentration inversion, thereby obtaining the updated sea ice concentration for the corresponding target area. This process is consistent with the inversion logic during the training of the sea ice concentration inversion model. The sea ice concentration inversion model calculates based on the characteristic information of the GNSS-R signal in the validation data, combined with the adjusted model parameters, and outputs the corresponding sea ice concentration value, i.e., the updated sea ice concentration, to reflect the actual inversion effect after iterative optimization of the model.

[0175] Subsequently, based on the updated sea ice concentration obtained from the inversion, a top-down view of the sea ice distribution in the target area is generated. This top-down view is a visual representation of the model inversion results. Typically, the sea ice distribution top-down view is centered on the polar region, presenting the spatial distribution characteristics of sea ice across the entire area from a direct overhead perspective. Different concentration areas are usually visually distinguished using color coding. For example, dark blue can represent low-density water areas (0%-15%), a gradient from light blue to white can represent ice-water transition zones (15%-85%), and pure white can represent high-density ice areas (above 85%), and so on.

[0176] Furthermore, by observing the spatial characteristics of the high-density ice zone in the top view, such as the shape of the ice edge, the direction of the ice edge, and the gradient changes in the transition zone, a direct comparison can be made with the actual sea ice concentration, and spatial deviations in the inversion results of the sea ice concentration inversion model can be quickly identified.

[0177] This visual verification method can intuitively confirm whether the updated sea ice concentration inversion model accurately captures the macroscopic distribution pattern and microscopic transition characteristics of sea ice, ultimately providing more comprehensive evidence for the reliability of the sea ice concentration inversion model.

[0178] As can be seen from the above, the solution provided in this application can capture the dynamic changes of sea ice over time by acquiring training data containing GNSS-R data and spatiotemporal alignment of real sea ice concentration at preset time intervals. Then, based on the GNSS-R data, the surface category of the target area is determined to be water or ice. Only the GNSS-R data of the ice surface is input into the sea ice concentration inversion model for inversion to obtain the predicted sea ice concentration. Furthermore, the model parameters of the sea ice concentration inversion model are iteratively adjusted by the loss value between the real sea ice concentration and the predicted sea ice concentration, so that the sea ice concentration inversion model focuses more on the feature learning of the ice surface GNSS-R data, reduces the interference of water surface GNSS-R data, thereby improving the ability to capture complex changes in sea ice, reducing the uncertainty and error of the inversion results, and improving the accuracy of sea ice concentration inversion.

[0179] like Figure 4 The diagram shown is a logical schematic of a GNSS-R-based sea ice concentration inversion model training method according to an embodiment of this application. The process includes:

[0180] Data Acquisition and Preprocessing: Training data consisted of GNSS-R data from FY-3E (including DDM) and real sea ice concentration data from OSI SAF, spanning from November 2023 to April 2024. Due to differences in spatiotemporal resolution between the two types of data, bilinear interpolation was used for spatiotemporal alignment with latitude and longitude of the reflection point and UTC time as constraints. Simultaneously, data quality control and filtering were performed, without specific limitations.

[0181] Feature extraction: The DDM data is extracted for features such as power ratio (PP), DDM average (DDMA), equivalent reflectivity (Γ), signal-to-noise ratio (SNR), kurtosis (K), and skewness (S). Simultaneously, quality control is performed, using SNR as a filtering criterion to remove DDM data with low SNR. Furthermore, based on the satellite identification numbers corresponding to the DDM data, the data is divided into three groups: GPS, BDS, and Galileo. These groups are used to train sea ice concentration inversion models for different navigation systems, reducing errors caused by mixed signals from different systems.

[0182] Sea ice detection: Based on historical sea ice concentration in ERA5, the monthly average sea ice concentration for each decade is determined. If the average sea ice concentration is less than 10% (first threshold), the surface category is water surface; if the average sea ice concentration is greater than 90% (second threshold), the surface category is ice surface. In the ice-water transition zone where the average sea ice concentration is between 10% and 90%, the number of pixels with power exceeding 30% of the peak value (preset ratio) is extracted from the DDM data as the effective pixel count N. If N is less than the preset threshold, the surface category is ice surface; if N is greater than or equal to the preset threshold, the surface category is water surface.

[0183] Model Building and Training: Radial basis function (RBF) SVR regression models were built for different navigation systems, using feature parameters of DDM data (land surface type as ice surface) as input and predicted sea ice concentration as output, serving as sea ice concentration inversion models. To adapt to interannual and seasonal variations in sea ice, a rolling training mechanism was adopted, updating training data every two weeks to further train and optimize the sea ice concentration inversion models.

[0184] Inversion and Validation Evaluation: Using the trained SVR model, sea ice concentration inversion was performed on FY-3E GNSS-R data from May 2023 to April 2024. GNSS-R data with actual sea ice concentrations between 15% and 85% were selected as validation data and input into the updated sea ice concentration inversion model to obtain the updated sea ice concentration for the target area. Furthermore, based on the updated and actual sea ice concentrations, the updated sea ice concentration inversion model can be validated and evaluated.

[0185] like Figure 5 As shown, the prediction accuracy can be evaluated by calculating the root mean square error (RMSE) and correlation coefficient (R). Furthermore, comparative analysis using top-down views of the polar distribution demonstrates that the inversion results exhibit excellent accuracy and consistency in both the Arctic and Antarctic regions. The inverted Arctic and Antarctic sea ice concentrations show good consistency with the actual concentrations. The average correlation coefficient (R) for the Arctic region is 0.9435, and the root mean square error (RMSE) is 0.1310. For the Antarctic region, the average correlation coefficient (R) is 0.9563, and the RMSE is 0.0946. In the more challenging ice-water transition zone, the RMSE values ​​remain within a reasonable error range, reaching 0.1700 for the Arctic and 0.1607 for the Antarctic.

[0186] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions involved are not necessarily required by this application.

[0187] Reference Figure 6 The diagram shows a structural block diagram of a GNSS-R-based sea ice concentration inversion model training device according to this application. The device may specifically include the following modules:

[0188] The acquisition module 201 is used to acquire training data at preset time intervals. The training data includes Global Navigation Satellite System Reflection (GNSS-R) data and real sea ice concentrations that are spatiotemporally aligned with the GNSS-R data.

[0189] Classification module 202 is used to determine the surface category of the corresponding target area based on the GNSS-R data, wherein the surface category is water surface or ice surface;

[0190] Prediction module 203 is used to input the GNSS-R data with the surface category of ice into the sea ice concentration inversion model to perform sea ice concentration inversion and obtain the predicted sea ice concentration of the corresponding target area.

[0191] The adjustment module 204 is used to determine the loss value between the actual sea ice concentration and the predicted sea ice concentration, and to iteratively adjust the model parameters of the sea ice concentration inversion model according to the loss value to obtain the updated sea ice concentration inversion model.

[0192] Optionally, the acquisition module 201 is specifically used for:

[0193] GNSS-R data and actual sea ice concentration are acquired at preset time intervals;

[0194] Based on latitude and longitude coordinates and time information, bilinear interpolation is used to perform spatiotemporal alignment of the GNSS-R data and the actual sea ice concentration to construct training data.

[0195] Optionally, the acquisition module 201 is specifically used for:

[0196] Acquire raw data, which includes sample GNSS-R data and sample real sea ice concentration;

[0197] A sliding time window is used to collect GNSS-R data and real sea ice concentration from the raw data. The data collection duration of each sliding time window is the preset time interval.

[0198] Optionally, the classification module 202 is specifically used for:

[0199] Obtain historical sea ice concentration, and based on the historical sea ice concentration, calculate the average sea ice concentration of the target area corresponding to the GNSS-R data within a preset time period;

[0200] If the average sea ice concentration is below a first threshold, the surface category of the target area is determined to be water surface;

[0201] If the average sea ice concentration is higher than the second threshold, the surface category of the target area is determined to be ice surface;

[0202] If the average sea ice concentration is higher than or equal to the first threshold and lower than or equal to the second threshold, the surface category of the corresponding target area is determined based on the GNSS-R data.

[0203] Optionally, the classification module 202 is specifically used for:

[0204] Extract the number of valid pixels from the GNSS-R data;

[0205] If the number of effective pixels is less than a preset threshold, the surface type of the target area is determined to be ice.

[0206] If the number of effective pixels is greater than or equal to the preset threshold, the surface category of the target area is determined to be water.

[0207] Optionally, the GNSS-R data includes a time-delay Doppler map, where each pixel in the time-delay Doppler map has a corresponding power; the classification module 202 is specifically used for:

[0208] Determine the power peak value in the time-delay Doppler plot;

[0209] The number of pixels whose power exceeds a preset percentage of the power peak value is determined as the effective pixel count.

[0210] Optionally, the GNSS-R data includes satellite numbers, and the prediction module 203 is specifically used for:

[0211] Based on the satellite number, the navigation system corresponding to each GNSS-R data with the surface category of ice is determined, resulting in multiple groups, each group including multiple GNSS-R data;

[0212] For each group, feature analysis is performed on each GNSS-R data within the group to obtain at least one corresponding feature parameter;

[0213] The feature parameters are input into the sea ice concentration inversion model corresponding to each group to perform sea ice concentration inversion and obtain the predicted sea ice concentration of the corresponding target area.

[0214] Optionally, the adjustment module 204 is specifically used for:

[0215] Determine the mean square error, mean absolute error, and correlation coefficient between the actual sea ice concentration and the predicted sea ice concentration;

[0216] The loss value is determined based on the mean square error, the mean absolute error, and the correlation coefficient.

[0217] Optionally, the device further includes a verification module for:

[0218] The GNSS-R data whose corresponding real sea ice concentration values ​​are within the preset range are used as verification data;

[0219] The verification data is input into the updated sea ice concentration inversion model to perform sea ice concentration inversion and obtain the updated sea ice concentration of the corresponding target area.

[0220] A top-down view of the sea ice distribution in the target area is generated based on the updated sea ice concentration. This top-down view is used to verify the updated sea ice concentration inversion model.

[0221] As can be seen from the above, the solution provided in this application can capture the dynamic changes of sea ice over time by acquiring training data containing GNSS-R data and spatiotemporal alignment of real sea ice concentration at preset time intervals. Then, based on the GNSS-R data, the surface category of the target area is determined to be water or ice. Only the GNSS-R data of the ice surface is input into the sea ice concentration inversion model for inversion to obtain the predicted sea ice concentration. Furthermore, the model parameters of the sea ice concentration inversion model are iteratively adjusted by the loss value between the real sea ice concentration and the predicted sea ice concentration, so that the sea ice concentration inversion model focuses more on the feature learning of the ice surface GNSS-R data, reduces the interference of water surface GNSS-R data, thereby improving the ability to capture complex changes in sea ice, reducing the uncertainty and error of the inversion results, and improving the accuracy of sea ice concentration inversion.

[0222] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0223] This invention also provides an electronic device, such as... Figure 7 As shown, it includes a processor 701, a communication interface 702, a memory 703, and a communication bus 704, wherein the processor 701, the communication interface 702, and the memory 703 communicate with each other through the communication bus 704.

[0224] Memory 703 is used to store computer programs;

[0225] When processor 701 executes a program stored in memory 703, it performs the following steps:

[0226] Training data is acquired at preset time intervals, including Global Navigation Satellite System Reflection GNSS-R data and real sea ice concentrations spatiotemporally aligned with the GNSS-R data.

[0227] Based on the GNSS-R data, the surface category of the corresponding target area is determined, which is either water or ice.

[0228] The GNSS-R data with the surface category of ice is input into the sea ice concentration inversion model to perform sea ice concentration inversion and obtain the predicted sea ice concentration of the corresponding target area.

[0229] The loss value between the actual sea ice concentration and the predicted sea ice concentration is determined, and the model parameters of the sea ice concentration inversion model are iteratively adjusted according to the loss value to obtain the updated sea ice concentration inversion model.

[0230] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0231] The communication interface is used for communication between the aforementioned terminal and other devices.

[0232] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0233] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0234] As can be seen from the above, in the solution provided in this application, by acquiring training data containing GNSS-R data and spatiotemporally aligned real sea ice concentration at preset time intervals, the dynamic changes of sea ice over time can be captured. Then, based on the GNSS-R data, the surface category of the target area is determined to be water or ice. Only the GNSS-R data of the ice surface is input into the sea ice concentration inversion model for inversion to obtain the predicted sea ice concentration. Furthermore, the model parameters of the sea ice concentration inversion model are iteratively adjusted by the loss value between the real sea ice concentration and the predicted sea ice concentration, so that the sea ice concentration inversion model focuses more on the feature learning of the ice surface GNSS-R data, reduces the interference of the water surface GNSS-R data, thereby improving the ability to capture complex changes in sea ice, reducing the uncertainty and error of the inversion results, and improving the accuracy of sea ice concentration inversion.

[0235] In another embodiment of the present invention, a computer-readable storage medium is also provided, which stores instructions that, when executed on a computer, cause the computer to perform a GNSS-R-based sea ice concentration inversion model training method as described in any of the above embodiments.

[0236] In another embodiment of the present invention, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute a GNSS-R-based sea ice concentration inversion model training method as described in any of the above embodiments.

[0237] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0238] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0239] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0240] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. A training method for a sea ice concentration inversion model based on GNSS-R, characterized in that, The method includes: Training data is acquired at preset time intervals, including Global Navigation Satellite System Reflection GNSS-R data and real sea ice concentrations spatiotemporally aligned with the GNSS-R data. Based on the GNSS-R data, the surface category of the corresponding target area is determined, which is either water or ice. The GNSS-R data with the surface category of ice is input into the sea ice concentration inversion model to perform sea ice concentration inversion and obtain the predicted sea ice concentration of the corresponding target area. Determine the loss value between the actual sea ice concentration and the predicted sea ice concentration, and iteratively adjust the model parameters of the sea ice concentration inversion model based on the loss value to obtain the updated sea ice concentration inversion model; The step of determining the surface category of the corresponding target area based on the GNSS-R data includes: Extract the number of valid pixels from the GNSS-R data; If the number of effective pixels is less than a preset threshold, the surface type of the target area is determined to be ice. If the number of effective pixels is greater than or equal to the preset threshold, the surface category of the target area is determined to be water surface; The GNSS-R data includes a time-delay Doppler map, where each pixel in the time-delay Doppler map has a corresponding power; extracting the effective pixel count from the GNSS-R data includes: Determine the power peak value in the time-delay Doppler plot; The number of pixels whose power exceeds a preset percentage of the power peak value is determined as the effective pixel count; The GNSS-R data includes satellite numbers. The process of inputting the GNSS-R data with a surface category of ice into a sea ice concentration inversion model to perform sea ice concentration inversion and obtain the predicted sea ice concentration for the corresponding target area includes: Based on the satellite number, the navigation system corresponding to each GNSS-R data with the surface category of ice is determined, resulting in multiple groups, each group including multiple GNSS-R data; For each group, feature analysis is performed on each GNSS-R data within the group to obtain at least one corresponding feature parameter; The feature parameters are input into the sea ice concentration inversion model corresponding to each group to perform sea ice concentration inversion and obtain the predicted sea ice concentration of the corresponding target area.

2. The method according to claim 1, characterized in that, The step of acquiring training data at preset time intervals includes: GNSS-R data and actual sea ice concentration are acquired at preset time intervals; Based on latitude and longitude coordinates and time information, bilinear interpolation is used to perform spatiotemporal alignment of the GNSS-R data and the actual sea ice concentration to construct training data.

3. The method according to claim 2, characterized in that, The acquisition of GNSS-R data and actual sea ice concentration at preset time intervals includes: Acquire raw data, which includes sample GNSS-R data and sample real sea ice concentration; A sliding time window is used to collect GNSS-R data and real sea ice concentration from the raw data. The data collection duration of each sliding time window is the preset time interval.

4. The method according to claim 1, characterized in that, The determination of the land surface category of the corresponding target area based on the GNSS-R data includes: Obtain historical sea ice concentration, and based on the historical sea ice concentration, calculate the average sea ice concentration of the target area corresponding to the GNSS-R data within a preset time period; If the average sea ice concentration is below a first threshold, the surface category of the target area is determined to be water surface; If the average sea ice concentration is higher than the second threshold, the surface category of the target area is determined to be ice surface; If the average sea ice concentration is higher than or equal to the first threshold and lower than or equal to the second threshold, the surface category of the corresponding target area is determined based on the GNSS-R data.

5. The method according to claim 1, characterized in that, Determining the loss value between the actual sea ice concentration and the predicted sea ice concentration includes: Determine the mean square error, mean absolute error, and correlation coefficient between the actual sea ice concentration and the predicted sea ice concentration; The loss value is determined based on the mean square error, the mean absolute error, and the correlation coefficient.

6. The method according to claim 1, characterized in that, After obtaining the updated sea ice concentration inversion model, the process further includes: The GNSS-R data whose corresponding real sea ice concentration values ​​are within the preset range are used as verification data; The verification data is input into the updated sea ice concentration inversion model to perform sea ice concentration inversion and obtain the updated sea ice concentration of the corresponding target area. A top-down view of the sea ice distribution in the target area is generated based on the updated sea ice concentration. This top-down view is used to verify the updated sea ice concentration inversion model.

7. A training device for a sea ice concentration inversion model based on GNSS-R, characterized in that, include: The acquisition module is used to acquire training data at preset time intervals. The training data includes Global Navigation Satellite System Reflection (GNSS-R) data and real sea ice concentrations that are spatiotemporally aligned with the GNSS-R data. The classification module is used to determine the surface category of the corresponding target area based on the GNSS-R data, wherein the surface category is water or ice. The prediction module is used to input the GNSS-R data with the surface category of ice into the sea ice concentration inversion model to perform sea ice concentration inversion and obtain the predicted sea ice concentration of the corresponding target area. An adjustment module is used to determine the loss value between the actual sea ice concentration and the predicted sea ice concentration, and to iteratively adjust the model parameters of the sea ice concentration inversion model based on the loss value to obtain the updated sea ice concentration inversion model. The classification module is specifically used for: Extract the number of valid pixels from the GNSS-R data; If the number of effective pixels is less than a preset threshold, the surface type of the target area is determined to be ice. If the number of effective pixels is greater than or equal to the preset threshold, the surface category of the target area is determined to be water surface; The GNSS-R data includes a time-delay Doppler map, where each pixel in the time-delay Doppler map has a corresponding power; the classification module is specifically used for: Determine the power peak value in the time-delay Doppler plot; The number of pixels whose power exceeds a preset percentage of the power peak value is determined as the effective pixel count; The GNSS-R data includes satellite identification numbers, and the prediction module is specifically used for: Based on the satellite number, the navigation system corresponding to each GNSS-R data with the surface category of ice is determined, resulting in multiple groups, each group including multiple GNSS-R data; For each group, feature analysis is performed on each GNSS-R data within the group to obtain at least one corresponding feature parameter; The feature parameters are input into the sea ice concentration inversion model corresponding to each group to perform sea ice concentration inversion and obtain the predicted sea ice concentration of the corresponding target area.

8. The apparatus according to claim 7, characterized in that, The acquisition module is specifically used for: GNSS-R data and actual sea ice concentration are acquired at preset time intervals; Based on latitude and longitude coordinates and time information, bilinear interpolation is used to perform spatiotemporal alignment of the GNSS-R data and the actual sea ice concentration to construct training data.

9. The apparatus according to claim 8, characterized in that, The acquisition module is specifically used for: Acquire raw data, which includes sample GNSS-R data and sample real sea ice concentration; A sliding time window is used to collect GNSS-R data and real sea ice concentration from the raw data. The data collection duration of each sliding time window is the preset time interval.

10. The apparatus according to claim 7, characterized in that, The classification module is specifically used for: Obtain historical sea ice concentration, and based on the historical sea ice concentration, calculate the average sea ice concentration of the target area corresponding to the GNSS-R data within a preset time period; If the average sea ice concentration is below a first threshold, the surface category of the target area is determined to be water surface; If the average sea ice concentration is higher than the second threshold, the surface category of the target area is determined to be ice surface; If the average sea ice concentration is higher than or equal to the first threshold and lower than or equal to the second threshold, the surface category of the corresponding target area is determined based on the GNSS-R data.

11. The apparatus according to claim 7, characterized in that, The adjustment module is specifically used for: Determine the mean square error, mean absolute error, and correlation coefficient between the actual sea ice concentration and the predicted sea ice concentration; The loss value is determined based on the mean square error, the mean absolute error, and the correlation coefficient.

12. The apparatus according to claim 7, characterized in that, The device further includes a verification module for: The GNSS-R data whose corresponding real sea ice concentration values ​​are within the preset range are used as verification data; The verification data is input into the updated sea ice concentration inversion model to perform sea ice concentration inversion and obtain the updated sea ice concentration of the corresponding target area. A top-down view of the sea ice distribution in the target area is generated based on the updated sea ice concentration. This top-down view is used to verify the updated sea ice concentration inversion model.

13. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of a GNSS-R-based sea ice concentration inversion model training method as described in any one of claims 1 to 6.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of a GNSS-R-based sea ice concentration inversion model training method as described in any one of claims 1 to 6.

Images