Sea ice concentration inversion model training method and device based on GNSS-R

Abstract

The invention provides a GNSS-R-based sea ice concentration inversion model training method and device, and the method comprises the steps: obtaining training data according to a preset time interval, and enabling the training data to comprise GNSS-R data and real sea ice concentration which is in time-space alignment with the GNSS-R data; determining an earth surface category of the corresponding target area based on the GNSS-R data, wherein the earth surface category is a water surface or an ice surface; inputting the GNSS-R data of which the earth surface category is the ice surface into a sea ice concentration inversion model to perform sea ice concentration inversion to obtain the predicted sea ice concentration of the corresponding target area; and determining a loss value between the real sea ice concentration and the predicted sea ice concentration, and performing iterative adjustment on the model parameters of the sea ice concentration inversion model according to the loss value to obtain an updated sea ice concentration inversion model. The sea ice concentration inversion model is more focused on feature learning of the GNSS-R data of the ice surface, and the accuracy of sea ice concentration inversion is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to a sea ice concentration inversion model training method and device based on GNSS-R. BACKGROUND

[0002] Sea ice concentration refers to the percentage of sea ice area in a specific sea area, which is not only an important input parameter for climate models, but also has important significance for the safety of the Arctic route.

[0003] In the prior art, 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 ground, the reflected signal of the L-band signal from the ground is received. Because sea water and sea ice have different reflection characteristics, the inversion model can analyze the characteristic differences of these reflected signals, and thus the sea ice concentration can be inverted.

[0004] However, the movement and change process of sea ice is very complex, especially the state of sea ice in the Antarctic region changes at any time, and the current inversion model cannot accurately capture these complex changes, resulting in great uncertainty and large error in the inversion result of sea ice concentration. SUMMARY

[0005] To solve the above technical problems, the present application shows a sea ice concentration inversion model training method and device based on GNSS-R, to solve the problem that the inversion model cannot accurately capture the complex changes of sea ice, resulting in great uncertainty and large error in the inversion result of sea ice concentration.

[0006] In a first aspect, the present application shows a sea ice concentration inversion model training method based on GNSS-R, comprising: Obtain training data at a predetermined time interval, wherein the training data includes GNSS-R (Global Navigation Satellite System Reflectometry) data and real sea ice concentration that is spatio-temporally aligned with the GNSS-R data; Determine the ground surface type of the corresponding target area based on the GNSS-R data, wherein the ground surface type is water surface or ice surface; Input the GNSS-R data with the ground surface type of ice surface into a sea ice concentration inversion model to invert the sea ice concentration, and obtain the predicted sea ice concentration of the corresponding target area; Determine the loss value between the real sea ice concentration and the predicted sea ice concentration, and 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.

[0007] Optionally, the acquiring the training data according to the preset time interval comprises: acquiring GNSS-R data and real sea ice density according to the preset time interval; based on the latitude and longitude coordinates and the time information, performing spatio-temporal alignment on the GNSS-R data and the real sea ice density by using bilinear interpolation to construct the training data.

[0008] Optionally, the acquiring the GNSS-R data and the real sea ice density according to the preset time interval comprises: acquiring original data, wherein the original data comprises sample GNSS-R data and sample real sea ice density; collecting GNSS-R data and real sea ice density from the original data by using a sliding time window, wherein the data collection time length of the sliding time window each time is the preset time interval.

[0009] Optionally, the determining the ground surface type of the corresponding target area based on the GNSS-R data comprises: acquiring historical sea ice density, and based on the historical sea ice density, counting the average sea ice density of the target area corresponding to the GNSS-R data within a preset time length; in the case that the average sea ice density is lower than a first threshold value, determining that the ground surface type of the target area is water surface; in the case that the average sea ice density is higher than a second threshold value, determining that the ground surface type of the target area is ice surface; in the case that the average sea ice density is higher than or equal to the first threshold value and lower than or equal to the second threshold value, determining the ground surface type of the corresponding target area according to the GNSS-R data.

[0010] Optionally, the determining the ground surface type of the corresponding target area according to the GNSS-R data comprises: extracting the number of effective pixels from the GNSS-R data; if the number of effective pixels is less than a preset threshold value, determining that the ground surface type of the target area is ice surface; if the number of effective pixels is greater than or equal to the preset threshold value, determining that the ground surface type of the target area is water surface.

[0011] Optionally, the GNSS-R data comprises a time delay Doppler diagram, each pixel in the time delay Doppler diagram has a corresponding power; and the extracting the number of effective pixels from the GNSS-R data comprises: determining the power peak value in the time delay Doppler diagram; The number of pixels whose power exceeds a preset percentage of the power peak value is determined as the effective pixel count.

[0012] 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: 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.

[0013] Optionally, 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.

[0014] Optionally, after obtaining the updated sea ice concentration inversion model, the method 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.

[0015] Secondly, this application discloses a training device for a GNSS-R-based sea ice concentration inversion model, comprising: 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. a prediction module, configured to input the GNSS-R data with the ground surface category of ice surface into a sea ice concentration inversion model to perform sea ice concentration inversion, and obtain predicted sea ice concentration of a target area corresponding to the GNSS-R data; an adjustment module, configured to determine a loss value between the real sea ice concentration and the predicted sea ice concentration, and iteratively adjust model parameters of the sea ice concentration inversion model according to the loss value, to obtain an updated sea ice concentration inversion model.

[0016] Optionally, the acquisition module is specifically configured to: acquire GNSS-R data and real sea ice concentration at preset time intervals; perform spatio-temporal alignment on the GNSS-R data and the real sea ice concentration based on longitude and latitude coordinates and time information by using bilinear interpolation, and construct training data.

[0017] Optionally, the acquisition module is specifically configured to: acquire original data, the original data including sample GNSS-R data and sample real sea ice concentration; collect GNSS-R data and real sea ice concentration from the original data by using a sliding time window, and a data collection time length of the sliding time window each time is the preset time interval.

[0018] Optionally, the classification module is specifically configured to: acquire historical sea ice concentration, and based on the historical sea ice concentration, count average sea ice concentration of a target area corresponding to the GNSS-R data within a preset time length; determine that a ground surface category of the target area is water surface in a case where the average sea ice concentration is lower than a first threshold value; determine that the ground surface category of the target area is ice surface in a case where the average sea ice concentration is higher than a second threshold value; determine the ground surface category of the target area corresponding to the GNSS-R data in a case where the average sea ice concentration is higher than or equal to the first threshold value and lower than or equal to the second threshold value.

[0019] Optionally, the classification module is specifically configured to: extract a number of effective pixels from the GNSS-R data; determine that the ground surface category of the target area is ice surface in a case where the number of effective pixels is less than a preset threshold value; determine that the ground surface category of the target area is water surface in a case where the number of effective pixels is greater than or equal to the preset threshold value.

[0020] Optionally, the GNSS-R data comprises a delay-Doppler map, each pixel in the delay-Doppler map has a corresponding power; the classification module is specifically configured to: determine a power peak in the delay-Doppler map; determine a number of pixels whose power exceeds a preset proportion of the power peak as the number of valid pixels.

[0021] Optionally, the GNSS-R data comprises a satellite number, and the prediction module is specifically configured to: based on the satellite number, determine a navigation system corresponding to GNSS-R data of each ground surface category as ice surface to obtain a plurality of groups, each group comprising a plurality of GNSS-R data; for each group, perform feature analysis on each GNSS-R data in the group to obtain at least one corresponding feature parameter; input the feature parameter into a sea ice density inversion model corresponding to each group to perform sea ice density inversion and obtain a predicted sea ice density of the target area.

[0022] Optionally, the adjustment module is specifically configured to: determine a mean square error, a mean absolute error and a correlation coefficient between the real sea ice density and the predicted sea ice density; based on the mean square error, the mean absolute error and the correlation coefficient, determine a loss value.

[0023] Optionally, the device further comprises a verification module configured to: input the GNSS-R data corresponding to the real sea ice density within a preset value range as verification data; input the verification data into the updated sea ice density inversion model to perform sea ice density inversion and obtain an updated sea ice density of the target area; generate a sea ice distribution overhead view of the target area according to the updated sea ice density, and the sea ice distribution overhead view is used to verify the updated sea ice density inversion model.

[0024] In a third aspect, the present application shows an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to realize the steps of the above-mentioned GNSS-R-based sea ice density inversion model training method.

[0025] In a fourth aspect, the present application shows a computer readable storage medium, and the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the GNSS-R-based sea ice concentration inversion model training method.

[0026] Compared with the prior art, the present application has the following advantages: In the present application, by acquiring training data containing GNSS-R data and spatio-temporally aligned real sea ice concentration at a preset time interval, the dynamic changes of sea ice over time can be captured. Then, based on the GNSS-R data, the ground surface type of the target area is determined as water surface or ice surface, and 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. Further, the model parameters of the sea ice concentration inversion model are iteratively adjusted by the loss value of 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 and reduces the interference of the water surface GNSS-R data, thereby improving the ability to capture the complex changes of sea ice, reducing the uncertainty and error of the inversion result, and improving the accuracy of sea ice concentration inversion. BRIEF DESCRIPTION OF DRAWINGS

[0027] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of the preferred embodiments and are not intended to limit the scope of the present application. Moreover, the same reference numerals are used throughout the various drawings to designate similar or equivalent parts. In the drawings: Figure 1 is a step flowchart of a GNSS-R-based sea ice concentration inversion model training method of the present application; Figure 2 is a distribution diagram of the correspondence between reflectivity and sea ice concentration in an embodiment of the present application; Figure 3 is a comparison diagram of the time delay Doppler diagrams corresponding to GPS, BDS and GALILEO respectively in an embodiment of the present application; Figure 4 is a logic diagram of a GNSS-R-based sea ice concentration inversion model training method in an embodiment of the present application; Figure 5 is a comparison diagram of the mean square error and the correlation coefficient calculated in an embodiment of the present application; Figure 6 is a structural block diagram of a GNSS-R-based sea ice concentration inversion model training device of the present application; Figure 7 is a structural block diagram of an electronic device of the present application. DETAILED DESCRIPTION

[0028] Exemplary embodiments of the present application will be described in detail with reference to the drawings. Although exemplary embodiments of the present application are shown in the drawings, it is understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thorough and complete, and will fully convey the scope of the application to those skilled in the art.

[0029] The sea ice concentration refers to the percentage of the area occupied by sea ice in a specific sea area. In the related art, the sea ice concentration can be inversed using the GNSS-R (Global Navigation Satellite System Reflectometry) technology. After the navigation satellite transmits an L-band signal to the ground, the reflected signal of the L-band signal from the ground is received. Since sea water and sea ice have different reflection characteristics, the sea ice concentration can be inversed by analyzing the characteristic differences of the reflected signals through an inversion model.

[0030] However, the movement and change process of sea ice is very complex, especially the state of sea ice in the Antarctic region changes at any time, and the current inversion model cannot accurately capture these complex changes, resulting in great uncertainty and large error in the inversed result of the sea ice concentration. Based on this, the GNSS-R-based sea ice concentration inversion model training method in the present application is proposed to solve the above problems.

[0031] The GNSS-R-based sea ice concentration inversion model training method provided by the embodiments of the present application will be described in detail below through specific embodiments.

[0032] Referring to Figure 1 , a step flowchart of the GNSS-R-based sea ice concentration inversion model training method of the present application is shown, which can specifically include the following steps: S11: Obtain training data according to a preset time interval, wherein the training data includes GNSS-R (Global Navigation Satellite System Reflectometry) data and real sea ice concentration that is spatiotemporally aligned with the GNSS-R data.

[0033] As Figure 2As shown, by analyzing the distribution map of the corresponding relationship between a certain characteristic parameter (reflectivity) of GNSS-R data in summer (such as August 2023) and winter (such as March 2024) and sea ice density, it can be obviously observed that there is a huge difference in spatial structure and numerical distribution, and this seasonal change brings a challenge to the generalization ability of the sea ice density retrieval model, resulting in that a GNSS-R-based sea ice density retrieval model trained based on training data of a single season is often difficult to adapt to the feature distribution of other seasons.

[0034] Based on this, in this step, the training data is continuously collected according to a preset time interval, aiming to capture the dynamic change characteristics of sea ice over time through sample coverage based on time series. Among them, the training data contains two parts: Global Navigation Satellite System Reflection (GNSS-R) data, mainly derived from observations of signal receiving equipment on ground-reflected signals, wherein the signal receiving equipment can be FY-3E GNOS-II (Global Navigation Satellite Occultation Sounder – II, Global Navigation Satellite Occultation Sounder – II), GNOS-II is a remote sensing instrument carried by the Chinese polar-orbiting meteorological satellite FY-3E, which supports the collection of reflected signals of signals emitted by navigation satellites of various systems such as GPS (Global Positioning System), BDS (BeiDou Navigation Satellite System), Galileo (Galileo Satellite Navigation System), etc. Real sea ice density data that is spatiotemporally aligned with GNSS-R data, wherein the real sea ice density data is usually derived from Level 3 sea ice density products provided by OSI SAF (Ocean and Sea Ice Satellite Application Facility).

[0035] In the training data, GNSS-R data and real sea ice density have completed spatiotemporal alignment, that is, each set of GNSS-R data can be accurately corresponded to real sea ice density of the same region and the same time, thereby providing a data basis with timeliness and spatial consistency for subsequent training of sea ice density retrieval models.

[0036] S12: Based on the GNSS-R data, determine the surface type of the corresponding target region, which is water surface or ice surface.

[0037] In this step, the significant difference in the reflection characteristics of L-band signals between seawater and sea ice can be utilized to achieve the classification of the ground surface category through in-depth analysis of GNSS-R data.

[0038] Specifically, the sea ice surface is relatively smooth, and the reflected signal energy is highly concentrated, with high peak power and narrow spatial range. In contrast, the seawater surface is usually rough, causing the reflected signal energy to be dispersed in space, with lower peak power and more obvious spatial expansion. Moreover, the equivalent reflectivity of seawater is usually lower than that of sea ice, while the signal-to-noise ratio may show the opposite trend due to the difference in surface roughness, and so on.

[0039] Based on the above physical differences, various algorithms can be employed to classify the ground surface category based on GNSS-R data, thereby filtering out GNSS-R data related to the ice surface and providing targeted input for the subsequent training of the sea ice density inversion model, improving the efficiency and accuracy of the training of the sea ice density inversion model.

[0040] S13: Inputting the GNSS-R data with the ground surface category as ice into the sea ice density inversion model to perform sea ice density inversion, and obtaining the predicted sea ice density corresponding to the target area.

[0041] In this step, the GNSS-R data with the ground surface category as ice can be inputted into the sea ice density inversion model to be trained, and the sea ice density inversion model can preliminarily predict the sea ice density of the target area and output the corresponding predicted sea ice density (range 0%-100%).

[0042] Among them, the sea ice density inversion model can be an SVR (Support Vector Regression) model, which adopts an RBF (Radial Basis Function kernel) kernel and can establish a nonlinear mapping between the GNSS-R data with the ground surface category as ice and the predicted sea ice density. Thus, after inputting the GNSS-R data with the ground surface category as ice into the sea ice density inversion model, the sea ice density inversion model can output the predicted sea ice density of the target area.

[0043] It can be understood that the core of sea ice density inversion is to inversely deduce the coverage ratio of the ice surface through the characteristics of GNSS-R data, and the reflection characteristics of water surface and ice surface are significantly different. If the GNSS-R data of the water surface is included in the training, the sea ice density inversion model may incorrectly learn the correlation between the water surface features and the sea ice density, resulting in a decrease in the inversion accuracy of the ice area.

[0044] Therefore, by focusing on the ice surface GNSS-R data, the sea ice concentration inversion model learns the inherent correlation between the ice surface GNSS-R data and the sea ice concentration, avoids the interference of the water surface GNSS-R data on the learning of the ice surface characteristics of the sea ice concentration inversion model, and thus strengthens the ability of the sea ice concentration inversion model to capture the ice surface reflection law.

[0045] S14: Determine the loss value between the real sea ice concentration and the predicted sea ice concentration, and iteratively adjust the model parameters of the sea ice concentration inversion model according to the loss value to obtain an updated sea ice concentration inversion model.

[0046] In this step, the model parameters can be iteratively adjusted by calculating the loss value between the real sea ice concentration and the predicted sea ice concentration, and finally an optimized sea ice concentration inversion model is obtained.

[0047] Specifically, the calculation of the loss value aims to quantify the deviation between the real sea ice concentration and the predicted sea ice concentration. If the real sea ice concentration and the predicted sea ice concentration differ greatly, a large loss value will be generated. Conversely, if the real sea ice concentration and the predicted sea ice concentration differ less, the loss value will be lower.

[0048] Based on the obtained loss value, the sea ice concentration inversion model can continuously fine-tune its model parameters (such as kernel function parameters, penalty coefficients, etc. in the support vector regression model) through the back propagation mechanism, gradually reduce the prediction deviation, and make the inversion accuracy of the sea ice concentration inversion model be improved in a targeted manner, and finally converge to an updated sea ice concentration inversion model that is more suitable for complex sea ice conditions.

[0049] In one implementation, the training data is obtained at a preset time interval, including: GNSS-R data and real sea ice concentration are obtained at a preset time interval; Based on the latitude and longitude coordinates and time information, the GNSS-R data and the real sea ice concentration are spatio-temporally aligned by bilinear interpolation to construct the training data.

[0050] In this implementation, first, GNSS-R data and real sea ice concentration are collected at preset time intervals. Each set of GNSS-R data corresponds to a specific "reflection point" on the ground, and the coordinates of the reflection point can be calculated through satellite orbit parameters, signal propagation path, etc. to accurately identify the spatial location of the signal source. Correspondingly, the real sea ice concentration is usually presented in a grid form, each grid cell corresponds to fixed latitude and longitude coordinates (such as 10km x 10km), and the geographical area covered by the latitude and longitude coordinates is clear. The time information of GNSS-R data and real sea ice concentration is directly related to the time of reception or update, reflecting the specific time of observation, wherein the time information can be based on UTC (Coordinated Universal Time) time, or other time bases, which are not limited.

[0051] Since there may be differences in spatial resolution or time granularity between the collection of GNSS-R data and real sea ice concentration data, it is necessary to construct training samples through spatio-temporal alignment processing.

[0052] Specifically, a bilinear interpolation method can be used to accurately align GNSS-R data and real sea ice concentration in space and time: in space, the grid information of the real sea ice concentration is mapped to the specific location of the GNSS-R reflection point through interpolation; in time, the UTC time is used for time dimension interpolation adjustment to ensure that the two correspond within the same spatio-temporal range. After bilinear interpolation, GNSS-R data and real sea ice concentration are accurately matched in time and space dimensions.

[0053] Further, quality control and screening of GNSS-R data and real sea ice concentration can also be performed, including removing outliers, data interpolation, denoising, normalization processing, etc., which are not limited.

[0054] Through this process, training data that meets the training needs of the sea ice concentration inversion model can be constructed, providing accurate matching input data for subsequent training of the sea ice concentration inversion model.

[0055] In one implementation, GNSS-R data and real sea ice concentration are obtained at preset time intervals, including: Obtaining original data, the original data including sample GNSS-R data and sample real sea ice concentration; Using a sliding time window to collect GNSS-R data and real sea ice concentration from the original data, the data collection time length of the sliding time window each time being the preset time interval.

[0056] In this implementation, the collection process of the training data relies on the sliding time window mechanism, and through the ordered extraction of the original data, it ensures that the training data can reflect the change characteristics of sea ice at different time scales.

[0057] First, the original data containing sample GNSS-R data and sample true sea ice density need to be collected in advance, wherein the sample GNSS-R data and the sample true sea ice density usually include multiple groups of data within a relatively long period of time.

[0058] In order to obtain training data at a preset time interval, a sliding time window is used to intercept the original data, and the single data collection duration of the sliding time window is set as the preset time interval (for example, half a month as a window). By moving the sliding time window on the time axis, GNSS-R data and corresponding true sea ice density within each sliding time window can be extracted from the original data each time.

[0059] This sliding mechanism can continuously and uniformly cover the entire time sequence of the original data, ensuring the continuity of the training data in time distribution, and ensuring the consistency of the time granularity of each batch of training data through the fixed window length, thereby providing support for the sea ice density inversion model to capture the short-term dynamic changes and long-term trends (such as ice evolution in seasonal replacement) of sea ice.

[0060] In an implementation, based on the GNSS-R data, the ground surface type of the corresponding target area is determined, including: The historical sea ice density is obtained, and based on the historical sea ice density, the average sea ice density of the target area corresponding to the GNSS-R data within a preset time length is counted; In the case where the average sea ice density is lower than a first threshold value, the ground surface type of the target area is determined as water surface; In the case where the average sea ice density is higher than a second threshold value, the ground surface type of the target area is determined as ice surface; In the case where the average sea ice density is higher than or equal to the first threshold value and lower than or equal to the second threshold value, the ground surface type of the corresponding target area is determined according to the GNSS-R data.

[0061] In this implementation, the historical sea ice density can be obtained first, and based on the historical sea ice density and the GNSS-R data, the ground surface type of the corresponding target area is determined, and it is clear that the target area belongs to “water surface” or “ice surface”.

[0062] The historical sea ice density refers to information such as the spatial distribution, coverage range and change trend of sea ice in a past period of time. For example, ERA5 (Fifth generation of the ECMWF Reanalysis for the global climate) sea ice coverage data can be used as the historical sea ice density. The ERA5 sea ice coverage data is generated by fusing satellite observations, buoy monitoring and other multi-source data, combined with numerical model simulation and data assimilation technology, and has the characteristics of wide global coverage, high data precision and complete time series.

[0063] Specifically, the average sea ice density of the target area in the preset time period can be calculated based on the historical sea ice density. The preset time period needs to be determined in combination with the periodicity of sea ice changes. For example, for sea areas with obvious seasonal changes, the historical sea ice density of the same period in the past 10 years (such as from December to the following March each year) can be selected to calculate the average value of the sea ice density of the target area in this period, that is, the average sea ice density.

[0064] In this way, by aggregating the time dimension of the historical sea ice density, the influence of short-term fluctuations can be weakened, and the average sea ice density is closer to the long-term stable state of the surface of the target area than the historical sea ice density at a single time.

[0065] Subsequently, the average sea ice density can be preliminarily determined according to the first threshold and the second threshold (wherein the first threshold is lower than the second threshold): when the average sea ice density is lower than the first threshold, it indicates that the target area is mainly characterized by open water in the preset time period, and therefore the target area is directly determined as water surface; when the average sea ice density is higher than the second threshold, it indicates that the target area is stably covered by sea ice for a long time, and the target area is determined as ice surface.

[0066] For example, assuming that the first threshold is 10% and the second threshold is 90%, if the average sea ice density of a certain area in January in the past 10 years is 95%, the area is usually ice surface in January, and if the average sea ice density is only 5%, the area is usually water surface in January.

[0067] For the area with an average sea ice density between the first threshold and the second threshold, it can be identified as an ice-water transition zone. Since the surface state of the ice-water transition zone is more affected by short-term changes (such as seasonal ice melting or freezing process), a single average sea ice density cannot accurately reflect the real-time properties, and therefore the surface class of the corresponding target area can be further determined according to the GNSS-R data.

[0068] This time-averaged classification method makes the determination of the surface type more objective and consistent, and can fully utilize the long-term rules contained in the historical sea ice concentration, providing a reliable basis for the subsequent GNSS-R data processing of the ice surface.

[0069] In an implementation, in step S12, the surface type of the corresponding target area is determined according to the GNSS-R data, including: extracting the number of effective pixels from the GNSS-R data; if the number of effective pixels is less than a preset threshold, determining that the surface type of the target area is ice surface; if the number of effective pixels is greater than or equal to the preset threshold, determining that the surface type of the target area is water surface.

[0070] In this implementation, the number of effective pixels N extracted from the GNSS-R data is introduced as an auxiliary judgment basis for the classification of the surface type. The number of effective pixels refers to the number of pixels in the GNSS-R data that can truly and reliably reflect the actual state of the surface, and can reflect the scattering characteristics of the surface, thereby distinguishing whether the ice-water transition zone is water surface or ice surface.

[0071] Specifically, the number of effective pixels and the scattering characteristics of the surface can be associated to further determine the surface type by a preset threshold: Generally, the scattering signal of the ice surface is more concentrated, and the number of effective pixels is less; the scattering signal of the water surface is more dispersed, and the number of effective pixels is more, so the number of effective pixels can be compared with the preset threshold, when the number of effective pixels is less than the preset threshold, the target area is divided into ice surface, otherwise it is water surface.

[0072] In this way, the efficiency of the determination 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 zone, and effectively improves the accuracy of the ice-water transition zone classification.

[0073] In an implementation, the GNSS-R data includes a delay-Doppler map, each pixel in the delay-Doppler map has a corresponding power; the number of effective pixels is extracted from the GNSS-R data, including: determining a power peak value in the delay-Doppler map; determining the number of pixels whose power exceeds a preset proportion of the power peak value as the number of effective pixels.

[0074] In this implementation, the extraction of the number of effective pixels and the determination of the surface type depend on the power distribution characteristics of the DDM (Delay-Doppler Map, delay-Doppler map), such as Figure 3The different navigation system DDM distribution diagrams are shown as examples, from left to right corresponding to GPS, BDS and GALILEO respectively. It can be seen that the high-power area (scattering area) shape, area and center position of the DDM of different navigation systems have significant differences, among which the DDM shape of the Galileo satellite is more concentrated, and the DDM of the GPS presents a larger area of the scattering area. Therefore, by quantifying the concentration degree of GNSS-R data scattering, the accurate differentiation of the properties of the target area can be realized, and the adaptability and generalization ability of the sea ice concentration inversion model to different satellite systems are improved.

[0075] Specifically, first, the power information is extracted from the DDM: all the power values of the pixels in the diagram are traversed, the maximum value among them is determined as the power peak value, and the total number of pixels whose power exceeds the power peak value by a preset proportion (such as 30%) is counted. This number is the number of effective pixels.

[0076] In this way, the physical characteristics of the DDM data are utilized to provide objective and reproducible basis for ground surface category determination.

[0077] In an implementation manner, the GNSS-R data includes satellite numbers, and the GNSS-R data with the ground surface category as ice is input into a sea ice concentration inversion model to perform sea ice concentration inversion, to obtain the predicted sea ice concentration of the corresponding target area, including: Based on the satellite numbers, the navigation system corresponding to each GNSS-R data with the ground surface category as ice is determined, to obtain multiple groups, each group including multiple GNSS-R data; For each group, feature analysis is performed on each GNSS-R data in the group, to obtain at least one corresponding feature parameter; The feature parameter is input into the sea ice concentration inversion model corresponding to each group to perform sea ice concentration inversion, to obtain the predicted sea ice concentration of the corresponding target area.

[0078] In this implementation manner, the inversion process of the sea ice concentration fully considers the signal characteristic differences of different navigation systems, and improves the prediction accuracy through grouping modeling.

[0079] Specifically, first, for the GNSS-R data with the ground surface category as ice, the GNSS-R data can be divided into navigation systems according to the satellite numbers contained therein: the satellite numbers are directly related to the navigation system to which they belong, and accordingly all GNSS-R data are divided into multiple independent groups, and the GNSS-R data in each group is from the same navigation system, to ensure that the GNSS-R data in the group has consistent signal transmission and reception characteristics. For each group, the GNSS-R data therein needs to be analyzed and extracted for features, and feature parameters reflecting the characteristics of the ground scattering are mined 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), skewness (S), etc. These feature parameters can quantify the differences in reflection after the interaction of different navigation system signals with sea ice and water surface.

[0080] Subsequently, for each group, a corresponding sea ice density inversion model (i.e., a dedicated model trained for different navigation systems) is matched, and the extracted feature parameters are input into the corresponding sea ice density inversion model for inversion calculation. Due to the inherent differences in signal frequency, coverage range, and scattering response of different navigation systems, the dedicated model can specifically fit the mapping relationship between the characteristics of each navigation system and the sea ice density, and finally output the predicted value of the sea ice density of each target area.

[0081] This way of modeling by navigation system effectively avoids the system error caused by the mixing of different system signals, makes the model more accurately capture the unique rules of each navigation system signal in sea ice inversion, and thus improves the reliability of the overall prediction.

[0082] In an implementation manner, the loss value between the real sea ice density and the predicted sea ice density is determined, including: determining the mean square error, the mean absolute error, and the correlation coefficient between the real sea ice density and the predicted sea ice density; determining the loss value based on the mean square error, the mean absolute error, and the correlation coefficient.

[0083] In this implementation manner, the determination of the loss value is through the comprehensive consideration of multi-dimensional error indicators, which comprehensively quantifies the deviation between the real sea ice density and the predicted sea ice density, and provides a more accurate direction for the optimization of the sea ice density inversion model.

[0084] Specifically, first, three types of error indicators can be calculated: one is the root mean squared error (RMSE), which is obtained by calculating the root value of the average of the square of the difference between the true sea ice density and the predicted sea ice density, which can intuitively reflect the average error size; the second is the mean absolute error (MAE), which is obtained by calculating the average of the absolute difference between the true sea ice density and the predicted sea ice density, which reflects the average level of overall error and is less affected by extreme values; the third is the correlation coefficient (R), which is used to evaluate the linear correlation between the true sea ice density and the predicted sea ice density. The closer the value of the correlation coefficient is to 1, the more consistent the trends of the two are.

[0085] After obtaining the above three indicators, they can be fused by a preset weight or fusion algorithm to obtain the loss value between the true sea ice density and the predicted sea ice density. For example, according to the influence of different indicators on the performance of the sea ice density inversion model, corresponding weights can be assigned, or after normalization processing, the indicators are mapped to the same order of magnitude and summed, etc. The specific implementation is not limited.

[0086] This multi-index fusion loss value calculation method not only avoids the one-sidedness of a single indicator, but also comprehensively reflects the performance of the prediction result in terms of error size, distribution rule and correlation, providing a more comprehensive basis for iterative adjustment of model parameters and promoting the optimization of the sea ice density inversion model to better fit the real surface state.

[0087] In one implementation, after obtaining the updated sea ice density inversion model, the method further includes: GNSS-R data corresponding to the true sea ice density value within the preset value range is taken as verification data; The verification data is input into the updated sea ice density inversion model for sea ice density inversion to obtain the updated sea ice density of the corresponding target area; The sea ice distribution plan view of the target area is generated according to the updated sea ice density, and the sea ice distribution plan view is used to verify the updated sea ice density inversion model.

[0088] In this implementation, the updated sea ice density inversion model needs to go through a further verification link to intuitively evaluate its ability to describe the sea ice distribution through spatial visualization means.

[0089] Specifically, first, GNSS-R data corresponding to real sea ice density values within a preset value range can be selected as verification data. The preset value range here can be set according to actual application requirements and key areas of sea ice monitoring. For example, focusing on the ice-water transition zone of 15%-85%, because the sea ice state in this area is complex and difficult to invert, the requirement for the accuracy of the sea ice density inversion model is higher, and selecting GNSS-R data in this range as verification data can more targetedly test the inversion ability of the sea ice density inversion model for complex sea ice states.

[0090] Subsequently, the verification data is input into the updated sea ice density inversion model for sea ice density inversion, thereby obtaining the updated sea ice density of the corresponding target area. This process is consistent with the inversion logic during the training of the sea ice density inversion model. The sea ice density inversion model will calculate according to the characteristic information of the GNSS-R signal in the verification data, combined with the adjusted model parameters, and output the corresponding sea ice density value, i.e., the updated sea ice density, to reflect the actual inversion effect of the model after iteration optimization.

[0091] Subsequently, based on the updated sea ice density obtained by inversion, a sea ice distribution overhead view of the target area is generated. The sea ice distribution overhead view is a direct presentation of the model inversion result. Generally, the sea ice distribution overhead view takes the polar region as the center and presents the spatial distribution characteristics of global sea ice from the overhead perspective. Different density areas are usually distinguished by color coding. For example, deep blue can represent low-density water area of 0%-15%, light blue to white can represent ice-water transition zone of 15%-85%, and pure white can represent high-density ice area of 85% or more.

[0092] Further, by observing the spatial characteristics of the high-density ice area, the trend of the ice edge line, and the gradient change of the transition zone in the overhead view, the real sea ice density can be directly compared, and the spatial deviation of the inversion result of the sea ice density inversion model can be quickly identified.

[0093] This visual verification method can confirm from an intuitive level whether the updated sea ice density inversion model accurately captures the macro distribution pattern and micro transition characteristics of sea ice, and ultimately provides more comprehensive evidence for the reliability of the sea ice density inversion model.

[0094] As can be seen from the above, in the scheme provided by the application, the training data containing GNSS-R data and spatio-temporally aligned real sea ice density is obtained at a preset time interval, which can capture the dynamic changes of sea ice over time. Then, based on the GNSS-R data, the ground surface type of the target area is determined as water surface or ice surface, only the GNSS-R data of the ice surface is input into the sea ice density inversion model for inversion to obtain the predicted sea ice density. Further, the model parameters of the sea ice density inversion model are iteratively adjusted based on the loss value of the real sea ice density and the predicted sea ice density, so that the sea ice density inversion model focuses more on the feature learning of the ice surface GNSS-R data and reduces the interference of the water surface GNSS-R data, thereby improving the ability to capture the complex changes of sea ice, reducing the uncertainty and error of the inversion result, and improving the accuracy of sea ice density inversion.

[0095] As shown in Figure 4 , it is a logic diagram of a GNSS-R-based sea ice density inversion model training method in an embodiment of the application, and the flow includes: Data acquisition and preprocessing: the training data is selected from the GNSS-R data (including DDM) of FY-3E and the real sea ice density from OSI SAF, and the time span is from November 2023 to April 2024. Since there is a difference in the spatio-temporal resolution of the above two kinds of data, the spatio-temporal alignment is performed by bilinear interpolation method with the reflection point longitude and latitude and UTC time as the constraints. At the same time, the data can also be subjected to quality control and screening, which is not limited.

[0096] Feature extraction: the feature parameters such as power ratio (PP), DDM average (DDMA), equivalent reflectivity (Γ), signal-to-noise ratio (SNR), kurtosis (K) and skewness (S) of DDM data are extracted. At the same time, quality control can be performed, and SNR is used as the screening condition to remove DDM data with low signal-to-noise ratio. In addition, according to the satellite number corresponding to the DDM data, the DDM data is divided into three groups of GPS, BDS and Galileo, which are respectively used to train the sea ice density inversion model for different navigation systems to reduce the error caused by mixed signals of different systems.

[0097] Sea ice detection: based on the historical sea ice density of ERA5, the monthly average sea ice density is determined every ten years. If the average sea ice density is less than 10% (the first threshold value), the ground surface type is water surface. If the average sea ice density is greater than 90% (the second threshold value), the ground surface type is ice surface. If the average sea ice density is between 10% and 90%, the ice-water transition zone, the number of pixels with power exceeding 30% (preset proportion) of the peak value is extracted from the DDM data as the effective pixel number N. If N is less than the preset threshold value, the ground surface type is ice surface. If N is greater than or equal to the preset threshold value, the ground surface type is water surface.

[0098] Model building and training: For different navigation systems, a radial basis kernel SVR regression model is built with the feature parameters of DDM data on ice surface as input and the predicted sea ice concentration as output, serving as the sea ice concentration inversion model. To adapt to interannual and seasonal changes of sea ice, a rolling training mechanism is adopted, and the training data is updated every half month to further train and optimize the sea ice concentration inversion model.

[0099] Inversion and verification evaluation: The trained SVR model is used to invert the sea ice concentration from May 2023 to April 2024 FY-3E GNSS-R data. Among them, the GNSS-R data corresponding to the true sea ice concentration of 15%-85% can be selected as the verification data, input into the updated sea ice concentration inversion model to invert the sea ice concentration, and the updated sea ice concentration of the target area is obtained. Further, based on the updated sea ice concentration and the true sea ice concentration, the verification and evaluation of the updated sea ice concentration inversion model can be realized.

[0100] As shown in Figure 5 , the prediction accuracy can be evaluated by calculating the root mean square error (RMSE) and the correlation coefficient (R). Further, combined with the polar distribution overhead view comparative analysis, it is proved that the inversion results in the Arctic and Antarctic show excellent precision and consistency. The inverted Arctic and Antarctic show good consistency with the true sea ice concentration, with an average correlation coefficient (R) of 0.9435 and a root mean square error (RMSE) of 0.1310 in the Arctic region, and an average correlation coefficient (R) of 0.9563 and a RMSE of 0.0946 in the Antarctic region. In the challenging ice-water transition zone, the RMSE value remains within a reasonable error range, reaching 0.1700 and 0.1607 in the Arctic and Antarctic, respectively.

[0101] It should be noted that for the method embodiments, in order to simply describe, they are all expressed as a series of action combinations, but those skilled in the art should know that the application is not limited by the order of the described actions, because according to the application, certain steps can be performed in other order or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification are all optional embodiments, and the actions involved are not necessarily required by the application.

[0102] Referring to Figure 6 , a structure block diagram of a GNSS-R-based sea ice concentration inversion model training device of the application is shown, which can specifically include the following modules: The acquisition module 201 is configured to acquire training data at a preset time interval, wherein the training data comprises global navigation satellite system reflection GNSS-R data and real sea ice density that is spatiotemporally aligned with the GNSS-R data; The classification module 202 is configured to determine a 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. The prediction module 203 is configured to input the GNSS-R data with the ground surface type of ice surface into a sea ice density inversion model to perform sea ice density inversion, so as to obtain predicted sea ice density of the corresponding target area. The adjustment module 204 is configured to determine a loss value between the real sea ice density and the predicted sea ice density, and iteratively adjust model parameters of the sea ice density inversion model according to the loss value, so as to obtain an updated sea ice density inversion model.

[0103] Optionally, the acquisition module 201 is specifically configured to: acquire GNSS-R data and real sea ice density at a preset time interval; spatiotemporally align the GNSS-R data and the real sea ice density based on longitude and latitude coordinates and time information by using bilinear interpolation, and construct training data.

[0104] Optionally, the acquisition module 201 is specifically configured to: acquire original data, wherein the original data comprises sample GNSS-R data and sample real sea ice density; collect GNSS-R data and real sea ice density from the original data by using a sliding time window, wherein a data collection time length of the sliding time window each time is the preset time interval.

[0105] Optionally, the classification module 202 is specifically configured to: acquire historical sea ice density, and statistically determine average sea ice density of a target area corresponding to the GNSS-R data within a preset time length based on the historical sea ice density; determine that a ground surface type of the target area is water surface in a case where the average sea ice density is lower than a first threshold value; determine that the ground surface type of the target area is ice surface in a case where the average sea ice density is higher than a second threshold value; determine the ground surface type of the corresponding target area based on the GNSS-R data in a case where the average sea ice density is higher than or equal to the first threshold value and lower than or equal to the second threshold value.

[0106] Optionally, the classification module 202 is specifically configured to: extracting an effective pixel number from the GNSS-R data; if the effective pixel number is less than a preset threshold, determining that a ground surface type of the target area is ice surface; if the effective pixel number is greater than or equal to the preset threshold, determining that the ground surface type of the target area is water surface.

[0107] Optionally, the GNSS-R data includes a delay-Doppler map, each pixel in the delay-Doppler map has a corresponding power; the classification module 202 is specifically configured to: determine a power peak value in the delay-Doppler map; determine a number of pixels whose power exceeds a preset proportion of the power peak value as the effective pixel number.

[0108] Optionally, the GNSS-R data includes a satellite number, and the prediction module 203 is specifically configured to: based on the satellite number, determine a navigation system corresponding to GNSS-R data in which each ground surface type is ice surface, to obtain a plurality of groups, each group including a plurality of GNSS-R data; for each group, performing feature analysis on each GNSS-R data in the group to obtain at least one corresponding feature parameter; inputting the feature parameter into a sea ice density inversion model corresponding to each group to perform sea ice density inversion, to obtain a predicted sea ice density of the target area corresponding thereto.

[0109] Optionally, the adjustment module 204 is specifically configured to: determine a mean square error, a mean absolute error, and a correlation coefficient between the true sea ice density and the predicted sea ice density; based on the mean square error, the mean absolute error, and the correlation coefficient, determine a loss value.

[0110] Optionally, the device further includes a verification module configured to: inputting the GNSS-R data corresponding to the true sea ice density value within a preset value range as verification data; inputting the verification data into the updated sea ice density inversion model to perform sea ice density inversion, to obtain an updated sea ice density of the target area corresponding thereto; generating a sea ice distribution overhead view of the target area according to the updated sea ice density, the sea ice distribution overhead view being used to verify the updated sea ice density inversion model.

[0111] It can be seen from the above that in the scheme provided in the application, the training data containing GNSS-R data and real sea ice density spatio-temporally aligned is obtained at a preset time interval, the dynamic change of sea ice over time can be captured, then the ground surface type of the target area is determined to be water surface or ice surface based on the GNSS-R data, only the GNSS-R data of the ice surface is input into the sea ice density inversion model for inversion to obtain the predicted sea ice density, and then the model parameters of the sea ice density inversion model are iteratively adjusted according to the loss value between the real sea ice density and the predicted sea ice density, so that the sea ice density inversion model focuses more on the feature learning of the ice surface GNSS-R data and reduces the interference of the water surface GNSS-R data, thereby improving the ability to capture the complex changes of sea ice, reducing the uncertainty and error of the inversion result, and improving the accuracy of sea ice density inversion.

[0112] For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant part can be referred to the part of the method embodiment.

[0113] The embodiment of the application further provides an electronic device, such as Figure 7 As shown in the figure, the electronic device comprises 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 complete mutual communication through the communication bus 704, The memory 703 is used for storing a computer program; The processor 701 is used for executing the program stored in the memory 703 to realize the following steps: Obtain training data at a preset time interval, wherein the training data comprises global navigation satellite system reflection GNSS-R data and real sea ice density spatio-temporally aligned with the GNSS-R data; Determine the ground surface type of the corresponding target area based on the GNSS-R data, wherein the ground surface type is water surface or ice surface; Input the GNSS-R data of the ice surface into a sea ice density inversion model for sea ice density inversion to obtain the predicted sea ice density of the corresponding target area; Determine the loss value between the real sea ice density and the predicted sea ice density, and iteratively adjust the model parameters of the sea ice density inversion model according to the loss value to obtain the updated sea ice density inversion model.

[0114] The communication bus mentioned by the terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0115] The communication interface is used for communication between the terminal and other devices.

[0116] The memory can include a Random Access Memory (RAM) and can also include a non-volatile memory, such as at least one disk memory. Optionally, the memory can also be at least one storage device located away from the aforementioned processor.

[0117] The processor mentioned above can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processing (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0118] As can be seen from the above, in the scheme provided in the present application, the training data containing GNSS-R data and spatio-temporally aligned real sea ice density is obtained at a preset time interval, which can capture the dynamic changes of sea ice over time. Then, based on the GNSS-R data, the ground surface category of the target area is determined as water surface or ice surface, and only the GNSS-R data of the ice surface is input into the sea ice density inversion model for inversion to obtain the predicted sea ice density. Further, the model parameters of the sea ice density inversion model are iteratively adjusted by the loss value of the real sea ice density and the predicted sea ice density, so that the sea ice density inversion model focuses more on the feature learning of the ice surface GNSS-R data and reduces the interference of the water surface GNSS-R data, thereby improving the ability to capture the complex changes of sea ice, reducing the uncertainty and error of the inversion result, and improving the accuracy of sea ice density inversion.

[0119] In a further embodiment provided by the present application, a computer readable storage medium is also provided, which stores instructions that, when executed on a computer, cause the computer to perform the GNSS-R based sea ice concentration retrieval model training method according to any one of the above embodiments.

[0120] In a further embodiment provided by the present application, a computer program product containing instructions is also provided, which, when executed on a computer, cause the computer to perform the GNSS-R based sea ice concentration retrieval model training method according to any one of the above embodiments.

[0121] In the above embodiments, the implementation can be totally or partially in software, hardware, firmware, or any combination thereof. When implemented in software, the implementation can be in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed by a computer, the computer instructions cause the computer to perform the processes or functions described in the embodiments of the present application. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable apparatus. The computer instructions can be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another computer readable storage medium, for example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) manner. The computer readable storage medium can be any available medium accessible by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be a magnetic medium (for example, floppy disk, hard disk, magnetic tape), an optical medium (for example, DVD), or a semiconductor medium (for example, Solid State Disk (SSD)) and the like.

[0122] It should be noted that, in this document, the terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, method, article or device including the element.

[0123] Each of the embodiments in the specification is described in a relevant manner, and the same or similar parts between the embodiments can be referred to each other. Each of the embodiments focuses on the difference from other embodiments. In particular, for the system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiments.

[0124] The above only describes the preferred embodiments of the present application, and is not used to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for training a GNSS-R based sea ice concentration retrieval model, characterized in that, The method comprises: acquiring training data at preset time intervals, wherein the training data comprises global navigation satellite system reflection GNSS-R data and real sea ice density spatiotemporally aligned with the GNSS-R data; determining the ground surface type of the 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 with the ground surface type being ice surface into a sea ice density inversion model to perform sea ice density inversion, thereby obtaining the predicted sea ice density of the corresponding target area; determining the loss value between the real sea ice density and the predicted sea ice density, and iteratively adjusting the model parameters of the sea ice density inversion model according to the loss value, thereby obtaining the updated sea ice density inversion model.

2. The method of claim 1, wherein, The acquiring of the training data at preset time intervals comprises: acquiring GNSS-R data and real sea ice density at preset time intervals; spatiotemporally aligning the GNSS-R data and the real sea ice density based on longitude and latitude coordinates and time information by using bilinear interpolation to construct the training data.

3. The method of claim 2, wherein, The acquiring of the GNSS-R data and the real sea ice density at preset time intervals comprises: acquiring original data, wherein the original data comprises sample GNSS-R data and sample real sea ice density; collecting GNSS-R data and real sea ice density from the original data by using a sliding time window, wherein the data collection time length of the sliding time window each time is the preset time interval.

4. The method of claim 1, wherein, The determination of the ground surface type of the corresponding target area based on the GNSS-R data comprises: acquiring historical sea ice density, and based on the historical sea ice density, counting the average sea ice density of the target area corresponding to the GNSS-R data within a preset time length; in the case that the average sea ice density is lower than a first threshold value, determining that the ground surface type of the target area is water surface; in the case that the average sea ice density is higher than a second threshold value, determining that the ground surface type of the target area is ice surface; in the case that the average sea ice density is higher than or equal to the first threshold value and lower than or equal to the second threshold value, determining the ground surface type of the corresponding target area according to the GNSS-R data.

5. The method of claim 1, wherein, The determination of the ground surface type of the corresponding target area based on the GNSS-R data comprises: extracting the number of effective pixels from the GNSS-R data; if the number of effective pixels is less than a preset threshold value, determining that the ground surface type of the target area is ice surface; if the number of effective pixels is greater than or equal to the preset threshold value, determining that the ground surface type of the target area is water surface.

6. The method of claim 5, wherein, The GNSS-R data comprises a time delay Doppler diagram, each pixel in the time delay Doppler diagram has a corresponding power; and the extraction of the number of effective pixels from the GNSS-R data comprises: determining the power peak value in the time delay Doppler diagram; determining the number of pixels whose power exceeds the preset proportion of the power peak value as the number of effective pixels.

7. The method of claim 1, wherein, The GNSS-R data includes a satellite number, the GNSS-R data of which the ground surface is classified as ice is input into a sea ice concentration inversion model to perform sea ice concentration inversion, and a corresponding predicted sea ice concentration of a target area is obtained, including: Based on the satellite number, a navigation system corresponding to each GNSS-R data of which the ground surface is classified as ice is determined, and a plurality of groups are obtained, each group including a plurality of GNSS-R data; For each group, feature analysis is performed on each GNSS-R data in the group to obtain at least one corresponding feature parameter; The feature parameters are input into a sea ice concentration inversion model corresponding to each group to perform sea ice concentration inversion, and a predicted sea ice concentration of a corresponding target area is obtained.

8. The method of claim 1, wherein, The loss value between the true sea ice concentration and the predicted sea ice concentration is determined, including: Determine the mean square error, mean absolute error and correlation coefficient between the true sea ice concentration and the predicted sea ice concentration; Based on the mean square error, the mean absolute error and the correlation coefficient, a loss value is determined.

9. The method of claim 1, wherein, After obtaining the updated sea ice concentration inversion model, further comprising: The GNSS-R data corresponding to the true sea ice concentration within a predetermined value range is used as validation data; The validation data is input into the updated sea ice concentration inversion model to perform sea ice concentration inversion, and an updated sea ice concentration of a corresponding target area is obtained; According to the updated sea ice concentration, a sea ice distribution overhead view of the target area is generated, and the sea ice distribution overhead view is used to verify the updated sea ice concentration inversion model. 10.A device for training a GNSS-R based sea ice concentration retrieval model, characterized in that, Comprising: An acquisition module is configured to acquire training data at a predetermined time interval, wherein the training data includes global navigation satellite system reflected GNSS-R data and true sea ice concentration that is spatio-temporally aligned with the GNSS-R data; A classification module is configured to determine the ground surface classification of a corresponding target area based on the GNSS-R data, wherein the ground surface classification is water surface or ice surface; A prediction module is configured to input the GNSS-R data of which the ground surface is classified as ice into a sea ice concentration inversion model to perform sea ice concentration inversion, and obtain a predicted sea ice concentration of a corresponding target area; An adjustment module is configured to determine the loss value between the true sea ice concentration and the predicted sea ice concentration, and iteratively adjust the model parameters of the sea ice concentration inversion model according to the loss value, to obtain an updated sea ice concentration inversion model.

11. The apparatus of claim 10, wherein, The acquisition module is specifically configured to: Acquire GNSS-R data and true sea ice concentration at a predetermined time interval; Based on longitude and latitude coordinates and time information, perform spatio-temporal alignment on the GNSS-R data and the true sea ice concentration using bilinear interpolation to construct training data.

12. The apparatus of claim 11, wherein, The acquisition module is specifically configured to: Acquire original data, wherein the original data includes sample GNSS-R data and sample true sea ice concentration; The GNSS-R data and the real sea ice density are collected from the original data by using a sliding time window, and each time of data collection of the sliding time window has a preset time interval.

13. The apparatus of claim 10, wherein, The classification module is specifically configured to: obtain historical sea ice densities, and based on the historical sea ice densities, count average sea ice densities of a target region corresponding to the GNSS-R data within a preset time length; determine that a ground surface type of the target region is water surface when the average sea ice density is lower than a first threshold value; determine that the ground surface type of the target region is ice surface when the average sea ice density is higher than a second threshold value; determine the ground surface type of the target region corresponding to the GNSS-R data when the average sea ice density is higher than or equal to the first threshold value and lower than or equal to the second threshold value.

14. The apparatus of claim 10, wherein, The classification module is specifically configured to: extract a number of valid pixels from the GNSS-R data; determine that the ground surface type of the target region is ice surface when the number of valid pixels is less than a preset threshold value; determine that the ground surface type of the target region is water surface when the number of valid pixels is greater than or equal to the preset threshold value.

15. The apparatus of claim 14, wherein, The GNSS-R data includes a time delay Doppler diagram, each pixel in the time delay Doppler diagram has a corresponding power; and the classification module is specifically configured to: determine a power peak value in the time delay Doppler diagram; determine a number of pixels whose power exceeds a preset proportion of the power peak value as a number of valid pixels.

16. The apparatus of claim 10, wherein, The GNSS-R data includes a satellite number, and the prediction module is specifically configured to: based on the satellite number, determine a navigation system corresponding to each GNSS-R data whose ground surface type is ice surface to obtain a plurality of groups, each group including a plurality of GNSS-R data; perform feature analysis on each GNSS-R data in each group to obtain at least one feature parameter corresponding to each GNSS-R data; input the feature parameter into a sea ice density inversion model corresponding to each group to perform sea ice density inversion and obtain a predicted sea ice density of a target region corresponding to each group.

17. The apparatus of claim 10, wherein, The adjustment module is specifically configured to: determine a mean square error, a mean absolute error, and a correlation coefficient between the real sea ice density and the predicted sea ice density; based on the mean square error, the mean absolute error, and the correlation coefficient, determine a loss value.

18. The apparatus of claim 10, wherein, The device further includes a verification module configured to: input the GNSS-R data corresponding to the real sea ice density within a preset value range as verification data; input the verification data into the updated sea ice density inversion model to perform sea ice density inversion and obtain an updated sea ice density of a target region corresponding to the verification data; generate a sea ice distribution overhead view of the target region based on the updated sea ice density, and the sea ice distribution overhead view is used to verify the updated sea ice density inversion model.

19. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the steps of the GNSS-R-based sea ice density inversion model training method according to any one of claims 1 to 9 when executing the program.

20. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the GNSS-R-based sea ice concentration inversion model training method in any one of claims 1 to 9.

Images