A method and system for regional quality assessment for polar sparse data

Abstract

The application provides a polar sparse data-oriented regional quality evaluation method and system, and belongs to the technical field of computer space data processing.The method comprises the following steps: obtaining polar space data, decoupling the characteristics of known sample points and unknown regional small units into polar space topology characteristics, observable shallow verification characteristics and to-be-predicted deep hidden characteristics; calculating the anisotropic equivalent space distance between the known sample points and the unknown regional small units based on the polar space topology characteristics, and performing screening in combination with background consistency constraints to construct a coarse screening candidate set; calculating the reconstruction residual of the coarse screening candidate set to the unknown regional small unit by using the observable shallow verification characteristics, performing update iteration according to the residual result until the convergence condition is met, and generating a fine sample set; and calculating the target hidden characteristic prediction column vector of the unknown regional small unit based on the fine sample set, and outputting the corresponding data quality confidence.

Description

Technical Field

[0001] This application relates to the field of computer spatial data processing technology, and more specifically, to a method and system for regional quality assessment of sparse polar data. Background Technology

[0002] As a cold source of the global climate system and a key region for geodynamics research, the accurate acquisition and analysis of polar spatial data plays an irreplaceable and fundamental role in climate change prediction, glacier dynamics simulation, and polar geological resource assessment. With advancements in remote sensing, airborne ice radar, and deep ice core drilling, the ability to acquire polar spatial data from multiple sources has improved. However, due to the extreme harshness of the natural environment, including polar nights, frigid temperatures, and blizzards, as well as the vastness of the polar ice sheets and the lack of conventional geographical control points, conducting in-situ physical surveys in the polar regions is extremely difficult and costly. These environmental constraints result in significant limitations in the acquisition of polar spatial data, leading to a highly heterogeneous and sparse spatial distribution of the obtained data.

[0003] Specifically, polar exploration data often exhibits linear or localized point-cluster clusters in spatial distribution. For example, airborne geophysical data is typically distributed continuously along fixed flight survey lines, while ground reconnaissance data is highly concentrated around specific research stations or along inland vehicle tracks. Between these survey lines and reconnaissance points lie vast gaps in physical exploration. Furthermore, polar spatial data possesses complex, multi-source, heterogeneous characteristics. It includes both shallow, continuous features easily obtained through satellite remote sensing, such as ice surface elevation, snow reflectivity, and surface temperature, and deep, hidden features that require heavy equipment and are extremely scarce, such as subglacial bedrock topography, total ice thickness, and bottom ice temperature. This data structure, characterized by relatively continuous shallow features and highly discrete deep features, presents significant challenges to the scientific assessment of the quality of polar spatial data across the entire region and the accurate extrapolation of attributes in unknown areas.

[0004] In existing spatial data quality assessment and regional characteristic extrapolation studies, for areas with a large number of data gaps, conventional spatial interpolation algorithms or traditional geostatistical models are typically used to fill in the data and assess its reliability. The core operational logic of these traditional methods is generally based on the first law of geography, which assumes that sampling points that are spatially closer have more similar attribute characteristics. Therefore, when processing spatial data, existing technologies mostly rely on Euclidean spatial distance as the absolute dominant standard for defining sample relevance and assigning evaluation weights, following the principle of nearest neighbor priority in sample matching and feature weighting.

[0005] However, directly applying the traditional assessment scheme relying on a single Euclidean distance and open-loop interpolation logic to polar environments reveals serious applicability flaws. Polar ice sheets and subglacial geological structures are profoundly influenced by complex glacial dynamics and ice flow topology, exhibiting strong anisotropy in spatial evolution. Two geographically close locations in polar regions, straddling ice dome watersheds or belonging to different flow velocity gradients, may have drastically different subglacial topography and deep geological features; conversely, points geographically distant along the same dynamic ice flow line often exhibit extremely high spatial correlation. Existing technologies, lacking consideration for specific polar dynamic topological constraints, easily include spuriously correlated samples with close physical distances but contradictory geological backgrounds in the sample selection stage. Furthermore, they lack a closed-loop self-validation mechanism for cross-validating the true correlation of samples based on known multidimensional features. This open-loop assessment model, which lacks constraints from the polar environment, is prone to introducing systematic biases when inferring high-value hidden features in unknown areas. Ultimately, this results in a serious lack of confidence in the spatial data quality assessment results, making it difficult to support the actual needs of modern high-precision polar scientific expeditions. Summary of the Invention

[0006] This invention provides a regional quality assessment method for polar sparse data, the method comprising: Acquire polar space data and decouple the features of known sample points and unknown small units into polar space topological features, observable shallow verification features, and deep hidden features to be predicted. Based on the topological features of polar space, the anisotropic equivalent spatial distance between known sample points and unknown small units is calculated, and a coarse candidate set is constructed by combining background consistency constraints. The reconstruction residuals of small units in unknown regions are calculated using observable shallow verification features. The update and iteration are performed based on the residual results until the convergence condition is met, and a fine sample set is generated. Calculate the target hidden feature prediction column vector based on the fine sample set for small units in the unknown region, and output the corresponding data quality confidence score.

[0007] Polar space topological features Includes three-dimensional coordinate vectors Vector of surface ice flow velocity tuples: ; in, Represents known sample points Its latitude, longitude, and absolute elevation; Characterizing known sample points Two-dimensional ice flow velocity vector projected onto a horizontal plane.

[0008] The calculation of anisotropic equivalent spatial distances between known sample points and unknown small units based on polar spatial topological features includes: By introducing the surface ice flow velocity direction vector, the physical distance is transformed into a distance that incorporates polar dynamics logic, and the known sample points are calculated. and unknown region small units Anisotropic equivalent spatial distance between : ; ; ; in, This is the streamline deviation penalty coefficient. The dynamic blocking penalty coefficient, Small units of unknown region The surface ice velocity vector at that location, This represents the highest historical ice surface flow velocity. For polar lateral stretching adjustment parameters The angle between the spatial line and the direction of the local ice flow. The spatial angle between the velocity vectors of ice flows at two points. The blocking constant of the polar watershed, The initial plane Euclidean distance is given.

[0009] By incorporating background consistency constraints, a coarse candidate set is constructed, which specifically includes: Using classification features as hard constraints, the known sample points are calculated using a discrete variable matching function. Small units of unknown regions Background consistency constraint coefficient : ; Set the preset equivalent distance threshold Extract samples that meet the distance threshold and have a background consistency constraint coefficient of 1 to construct an initial coarse candidate set. : ; in, , These represent polar ice facies zone types and subglacial landform types, respectively. Representing small units of unknown regions The corresponding polar ice facies types and subglacial landform types; This is a function for matching discrete variables.

[0010] The reconstruction residuals of small units in unknown regions are calculated using observable shallow verification features, specifically including: Utilizing small units of unknown regions Using real observation data as anchor points, the current iteration step is driven. The coarse candidate set below Perform reverse physical quantity reconstruction and calculate the reconstructed shallow feature vector. : ; ; Calculate and reconstruct shallow feature vectors With real, observable, shallow verification feature vectors Reconstruction residuals between : ; in, For anisotropic equivalent spatial distance, For the initial reconstruction weights, The polar space attenuation index; For known sample points The observable shallow verification feature vector.

[0011] The update iteration is performed based on the residual results, including steps such as eliminating spurious correlation samples and using boundary samples for collaborative gain. Specifically, this involves: processing known sample points... Temporarily remove a sample from the set and calculate the reconstruction residual after the sample is missing. and define known sample points Error contribution sensitivity : Introducing a suboptimal boundary sample pool Suboptimal boundary sample pool For edge points, the known sample points are those whose equivalent distance is greater than a preset equivalent distance threshold and meet the background consistency condition; Simulate adding it to the current set and calculate the newly added reconstruction residual after addition. Define edge points The introduction of gain : .

[0012] The execution of the update iteration further includes the step of performing a bidirectional update of the set based on error contribution sensitivity and the introduction of gain: Extracting the sensitivity of the maximum error contribution in the current iteration step and their corresponding candidate points to be eliminated and maximum introduced gain and their corresponding candidate points to be introduced ; like and This indicates that removing spurious samples yields higher returns, so the removal operation is performed, and the candidate points to be removed are... Remove from the current collection; like and This indicates that introducing edge samples yields higher returns. Therefore, the introduction operation is performed, adding candidate points to be introduced. Add it to the current set and permanently remove it from the suboptimal boundary sample pool.

[0013] Until the convergence condition is met, a refined sample set is generated, which specifically includes: Repeat the forward reconstruction, residual comparison, sensitivity analysis, and dynamic update process until any of the following convergence conditions are triggered. The final stable candidate set is then output as the refined sample set. Condition 1: Reconstructing the residual Less than the polar tolerance threshold Condition 2: Sensitivity to maximum error contribution and maximum introduced gain All are less than or equal to zero; Condition 3: The total number of remaining samples in the set reaches the extremely low limit for maintaining the degrees of freedom of polar space interpolation. .

[0014] The calculation of target concealment feature prediction column vectors for small units in unknown regions based on a refined sample set specifically includes: based on anisotropic equivalent spatial distance Calculate the fine sample set Known sample points For small units in unknown regions Final prediction weights : Combined with the final prediction weights For known sample points The column vector of deep hidden features to be predicted By performing weighted fusion, small units of the unknown region can be inferred. Target hidden feature prediction column vector : in, This is the preset deep feature space attenuation coefficient.

[0015] This invention also provides a regional quality assessment system for polar sparse data, the system comprising: Feature acquisition module: acquires polar space data and decouples the features of known sample points and unknown small units into polar space topological features, observable shallow verification features, and deep hidden features to be predicted; Coarse candidate set construction module: Based on the polar spatial topological features, calculate the anisotropic equivalent spatial distance between known sample points and unknown small units, and combine it with background consistency constraints to screen and construct a coarse candidate set; The fine sample set construction module calculates the reconstruction residual of the coarse candidate set for small units in the unknown region using observable shallow verification features, and performs updates and iterations based on the residual results until the convergence condition is met, thus generating the fine sample set. Evaluation module: Calculates the target hidden feature prediction column vector of small units in unknown regions based on a fine sample set, and outputs the corresponding data quality confidence score.

[0016] This application provides a regional quality assessment method and system for sparse polar data, effectively overcoming the technical bottleneck of highly scarce and spatially unevenly distributed physical survey data in extreme polar environments. This application abandons the inherent logic of traditional geostatistics that relies solely on Euclidean physical distance for interpolation. Instead, it decouples multi-source, heterogeneous polar measured data into three independent dimensions: spatial dynamic topology, observable shallow surface conditions, and deep, hidden attributes. It introduces the polar-specific surface velocity vector and the large-scale watershed topology of the polar ice sheet into spatial distance measurements, while simultaneously integrating ice facies zones and subglacial landforms as rigid category constraints. This constraint mechanism can accurately identify and forcibly intercept interfering samples that are physically close but located in different lateral shear strain zones or have opposing material transport directions.

[0017] To eliminate spurious correlation noise hidden by local deep dynamic anomalies, this application introduces a residual iterative screening method based on inversion reconstruction of observable features. Utilizing readily available, continuous, shallow surface observation data obtained through space-based remote sensing as real information for multidimensional feature verification, the system drives the inverse reconstruction and inference of surface physical quantities from the candidate sample set formed in the initial screening stage. Based on the residual gradient distribution between the reconstructed features and the actual observed features, the system dynamically removes contaminated samples that deteriorate the fitting of local features and adaptively introduces boundary samples with potential for physical synergistic gains along the direction of reducing residuals. This forces the associated samples participating in the evaluation to complete verification on known surface physical dimensions, ensuring that the final reference sample set shares the same set of physical evolution laws with the unknown polar regions.

[0018] This application not only outputs high-precision numerical projections of deep subglacial concealment features, but also constructs a multi-dimensional quality evaluation model to quantify the reliability of the projection results. By comprehensively evaluating the final fitting convergence level of the iterative closed loop, the effective support density of high-quality spatial samples, and the physical variability of local concealment features, this application simultaneously generates a spatial data quality confidence index for the prediction results. Attached Figure Description

[0019] Figure 1 This is a flowchart of the regional quality assessment for sparse polar data according to the present invention; Figure 2 A sensitivity distribution diagram of error contribution for known samples in this invention; Figure 3This is an iterative graph based on shallow features. Detailed Implementation

[0020] This embodiment provides a regional quality assessment system for sparse polar data, which is applied to the detection of polar ice sheets, ice shelves and their internal glacier dynamics and subglacial geological structures.

[0021] The regional quality assessment system for sparse polar data includes a multi-source polar data acquisition subsystem, a polar communication and data storage subsystem, a computational assessment subsystem, and a data visualization output subsystem.

[0022] The multi-source polar data acquisition subsystem is used to acquire measured spatial data in the polar natural environment. Based on the special physical conditions of polar exploration, the hardware equipment of the multi-source polar data acquisition subsystem specifically includes an air-based detection platform, a space-based remote sensing platform, and a ground-based survey platform.

[0023] Preferably, the space-based remote sensing platform includes a cluster of remote sensing satellites operating in a polar sun-synchronous orbit. The satellite payloads are equipped with altimeters, synthetic aperture radars, and multispectral imagers, and are configured to continuously acquire surface physical parameters of the entire polar region, specifically for collecting large-scale data on ice surface elevation, ice and snow cover surface temperature, and reflectivity.

[0024] Preferably, the airborne detection platform includes a polar fixed-wing aircraft, with an airborne ice-detecting radar system mounted on its fuselage or wings. The fixed-wing aircraft is configured to fly along a preset polar survey grid, and the airborne ice-detecting radar system transmits and receives electromagnetic wave signals that penetrate the ice layer, specifically used to collect discretely distributed data on the thickness of deep ice layers, the elevation profile of bedrock beneath the ice, and isochronous interface data within the ice layer.

[0025] Preferably, the surface-based survey platform includes deep ice core drilling units and ice surface deployment arrays deployed along the polar research station route and on typical cross-sections in the polar inland. The ice surface deployment array includes multiple GNSS base station devices equipped with independent, cryogenic power supplies. These GNSS base station devices are fixed to the ice surface as beacons, and by continuously tracking the absolute three-dimensional coordinate displacement of the beacons over a long period, they are specifically used to collect absolute coordinate reference data and surface ice flow velocity direction and velocity vector data.

[0026] The deep ice core drilling unit is deployed in the main polar ice body or ice dome area, and includes deep ice core electromechanical drilling tools and temperature measuring probes. Through mechanical core sampling with the drill bit, absolute ice bottom lithological physical samples are obtained at the actual deep borehole locations. Simultaneously, using a distributed temperature sensor array or in-hole temperature measuring probes deployed along the borehole axis, the measured ice temperature data at the bottom rock contact surface is directly detected and recorded. Drilling operations to obtain in-situ physical data of the bottom layer are existing engineering technologies in the field of polar exploration. Drilling operations can refer to the European Antarctic Ice Core Drilling Program (EPICA) and the Chinese Antarctic Kunlun Station deep ice core drilling project.

[0027] The computational evaluation subsystem is the processing hub of this system, and preferably includes a server workstation with a parallel computing architecture. This server workstation comprises multiple microprocessor arrays and non-volatile computer-readable storage media. The storage media stores a polar space data region quality evaluation program, and the microprocessor arrays are configured to execute the evaluation program. The data visualization output subsystem includes a high-resolution graphics display terminal configured to receive the processing results from the server workstation and render a polar space data inference layer. This layer synchronously overlays and displays the inferred deep concealed feature contour lines in a two-dimensional plane or three-dimensional space, as well as a transparency gradient confidence heatmap rendered based on the data reliability comprehensive score.

[0028] Based on the hardware architecture and data aggregation of the aforementioned multi-source polar data acquisition subsystem and polar communication and data storage subsystem, this embodiment details the specific steps of the server workstation executing the polar spatial data area quality assessment program. The assessment program first extracts multi-dimensional polar features, and the specific implementation process is as follows: The server workstation reads the entire polar region's exploration data from the relational spatial database in the local data center and performs spatial discretization and feature heterogeneity decoupling operations. Due to the harsh polar environment, the distribution of known measured points is extremely uneven, and there are fundamental differences in attributes, dimensions, and acquisition density between the continuous surface data acquired by satellite remote sensing and the sparse deep data acquired by drilling / radar. If all features are indiscriminately mixed into a traditional distance-weighted model, high-frequency surface noise will overwhelm low-frequency deep geological features. Therefore, this step decouples all prior spatial data into three subsets with independent features.

[0029] The processor divides the unknown region of the polar target to be evaluated into several regular two-dimensional or three-dimensional spatial grids, defined as a set of small units of the unknown region. Its mathematical expression is: ; in, Indicates the first A small unit of unknown region, This represents the total number of small units in the unknown region obtained by dividing the area.

[0030] Meanwhile, the geographic coordinates of points from which complete deep-seated measurement data has been obtained through field surveys or ice-penetrating radar are defined as known sample points. The size of the set of known sample points is denoted as . ,in .

[0031] For any known sample point The processor extracts the associated multidimensional attributes from the database and decouples them into tripartite feature sets. Defined as: ; Simultaneously, for any unknown small unit Due to the lack of deep measured data, its characteristic set is formally expressed as: ; Polar space topological features The spatial topological features (or This includes traditional Euclidean coordinates while also incorporating the anisotropic rheological characteristics of polar glaciers. Polar ice sheets are not static entities but rather non-Newtonian fluids with rheological properties; the similarity of subglacial geological properties is strictly controlled by the material transport patterns of ice flows. Therefore, spatial topological characteristics... Defined as containing three-dimensional coordinate vectors Vector of surface ice flow velocity tuples: ; in, Representing known sample points respectively Its latitude, longitude, and absolute elevation; This characterizes the two-dimensional ice flow velocity vector projected onto the horizontal plane from a point obtained by ice surface array measurements or satellite interferometric radar measurements. This topological feature... Used to construct anisotropic spatial constraint boundaries based on ice streamlines.

[0032] Observable shallow verification features The observable shallow verification features (or These are surface physical parameters that can be acquired over a large area and continuously via space-based remote sensing platforms. The characteristics of observable shallow surface verification features are low acquisition cost, high spatial coverage, and availability at known sample points. and unknown region small units All locations have complete actual observation values. Expressed as Dimensional column vectors: ; in, This represents the total number of dimensions of the observable shallow validation features. Represents known sample points In the Initial measurements on an observable feature.

[0033] As a preferred embodiment of this application for the polar ice sheet exploration scenario, the total dimension of observable shallow surface verification features is set. Under this preferred setting, column vectors Each dimension in the equation corresponds to the following polar-specific physical quantities: First dimension The absolute elevation of the ice sheet surface, acquired by a polar-orbiting laser altimeter, is used to characterize the macroscopic three-dimensional geometry of the ice body; the second dimension... The third dimension is defined as the surface slope of the ice sheet, calculated based on the surface elevation, and is used to characterize the gravitational potential energy gradient that drives ice rheology. The fourth dimension is defined as the surface albedo of ice and snow, acquired by the polar multispectral imaging payload, used to characterize the physical state and radiation budget characteristics of polar surface ice and snow; The polar microwave surface brightness temperature was set and acquired by a spaceborne microwave radiometer. Signals in this frequency band have a certain penetration depth into shallow polar ice layers, and are used to characterize the volume scattering properties of the shallow ice layer; the fifth dimension. The ice fissure density index is set as the transverse shear strain intensity of ice flow in the region, based on the texture feature extraction of synthetic aperture radar images.

[0034] Considering that polar ice surfaces often exhibit extreme outliers caused by large crevasses or localized windblown snow, conventional mean-variance standardization can skew polar background features due to these outliers. Therefore, this embodiment introduces a standardization algorithm based on absolute median difference. and Perform dimensional unification. For the first... Each feature dimension, its absolute median difference and standardized eigenvalues The calculation model is as follows: ; ; The above mapping eliminates the dimensional differences between different remote sensing physical quantities, generating a standardized verification feature set. It is not used as a direct prediction target in the whole system, but as verification information during iteration, for backtesting and correcting the true correlation of the known sample set.

[0035] Deep concealed features to be predicted The deep hidden features to be predicted These are subglacial properties obtained through airborne ice-penetrating radar or deep ice core drilling. Acquiring this type of data is extremely difficult, and is only possible with a limited number of known sample points. Actual measured values ​​exist in certain locations, while in unknown small units... The state is an unknown state to be solved (denoted as ). ). Expressed as Dimensional column vectors: ; in, This represents the total dimension of the deep hidden features to be predicted. Represents known sample points In the Measured values ​​on concealed features. Concealed features It is the predictive target variable for regional quality assessment and spatial characteristic deduction in this system.

[0036] As a preferred embodiment of this application, the total dimension of the deep hidden features to be predicted is set. Under this preferred setting, column vectors Each dimension in the equation corresponds to the following physical quantities in the deep polar regions: First dimension Set as the total thickness of the ice sheet's depth, second dimension Set as the temperature of the bottom ice-rock interface, third dimension Set to the electromagnetic wave reflection intensity of the substrate. After completing the multidimensional feature extraction of the polar regions, this system obtains feature vectors. Existing conventional geostatistical interpolation generally assumes that the space is isotropic, that is, it relies solely on physical Euclidean distance to determine sample similarity. However, the polar ice cap has formed a highly complex glacial dynamic network during its long evolution driven by gravity. In the polar scenario, two spatial points located upstream and downstream of the same ice flow line, even if physically far apart, are highly correlated in terms of subglacial topography and evolutionary history; conversely, if two spatial points are physically close, crossing an ice dome watershed or belonging to ice facies zones with opposing flow directions, their deep, hidden characteristics are often quite different.

[0037] Based on the aforementioned polar constraints, the server workstation performs anisotropic candidate set construction based on polar features. This embodiment first constructs an anisotropic space penalty function.

[0038] Extracted polar dynamic spatial topological features (For known sample points) )and (For small units in unknown areas) Based on the direction vector of ice surface flow velocity. This transforms physical distance into anisotropic equivalent distance that incorporates polar dynamics logic, thus introducing a spatial penalty.

[0039] Three-dimensional coordinate vector and Project onto the stereographic projection plane and obtain the corresponding two-dimensional plane coordinates. and And calculate the small units of the unknown region. Pointing to known sample points Planar direction vector and the initial plane Euclidean distance : ; ; The processor uses the surface ice flow velocity direction vector in the spatial topology to calculate two polar space penalty coefficients: streamline deviation penalty coefficient. With dynamic blocking penalty coefficient .

[0040] Streamline deviation penalty coefficient The weakening effect of transverse shear zones of polar ice flows on geological correlations was considered. Data correlation along the ice flow direction (longitudinal) is significantly greater than that perpendicular to the ice flow direction (transverse). The processor calculates the planar direction vector. and unknown region small units Surface ice velocity vector at the location The cosine of the angle between them is used to construct the streamline deviation penalty coefficient: ; ; in, Indicates the angle between the spatial line and the direction of the local ice flow; This represents the historical maximum ice surface velocity value within the polar research area, used to normalize the velocity. The preset polar lateral stretching adjustment parameters (preferred setting range is) The physical logic of the above formula lies in: when the sample points are known... With unknown units When the line connecting them is completely parallel to the direction of the ice flow (i.e.) No streamline deviation penalty ( The physical distance is not magnified; however, when the line connecting the two points is perpendicular to the direction of the ice flow (crossing the ice flow), the penalty coefficient reaches its maximum, and the local ice flow velocity is also increased. The faster the shearing, the stronger the physical isolation caused by the lateral shearing, resulting in the physical distance being magnified many times over.

[0041] Dynamic blocking penalty coefficient If two adjacent points are located on opposite sides of a watershed, the directions of ice material transport will be completely opposite, lacking the geological basis for property transfer. (Processor extraction) and velocity vector at the location and Calculate the consistency of the flow direction between the two: ; ; in, This represents the spatial angle between the velocity vectors of ice flows at two points; The preferred value is the preset polar watershed blocking constant. .

[0042] If the ice flows in the same direction at both points ( If so, it means they belong to the same ice basin, and the penalty coefficient is... If the flow velocities at two points are in opposite directions (e.g., crossing a watershed), The exponential penalty function will increase non-linearly, thus logically forcibly pushing these two spurious correlation points, which are physically close, infinitely far apart.

[0043] Taking into account the above polar dynamic topological constraints, the processor ultimately calculates the Euclidean distance. Mapped to anisotropic equivalent spatial distance : ; By constructing an anisotropic space penalty function, this system eliminates pseudo-nearest neighbors that are physically close but have contradictory dynamic logic in the special polar environment. The equivalent distance calculated at this point is... It is not simply a spatial scale, but a geologically related distance that incorporates the topological relationships of polar ice flows.

[0044] The polar ice caps cover a vast area, spanning drastically different physical and geological environments. In the shallow polar layers, the ice mass can be divided into vastly different glacial facies zones, such as dry snow zones, permeable zones, and bare ice zones. In the deep polar regions, the subsurface rock structures also exhibit distinctly different landform types, including subglacial highlands, subglacial troughs, and subglacial lake basins. If two spatial points belong to different glacial facies zones or geomorphic units, the evolutionary mechanisms of their hidden deep features will be fundamentally altered. Therefore, even with anisotropic equivalent spatial distances... If the sample size is too small, it cannot be directly considered valid; macro-level classification characteristics must be used as a hard constraint.

[0045] The processor constructs known sample points based on the extracted multidimensional features and the associated polar geological base map data. and unknown region small units The polar discrete eigenvectors are denoted as follows: and : in, Represents known sample points The type of polar ice facies in which it is located, such as dry snow zone or bare ice zone; Represents known sample points The subglacial landform type, such as subglacial canyon type, uplifted highland type, etc.

[0046] Define the background consistency constraint coefficient Calculate using the discrete variable matching function (Kronecker function) and The degree of matching on the above discrete features: Among them, discrete variable matching function The logical definition is: if the macro categories are the same, then... ,but If the macro categories are different, then... ,but .

[0047] Only when the sample points are known With unknown units When the glacial facies type and the subglacial landform type are completely identical Only equals 1; as long as there is any inconsistency in the background, for example Located in the subglacial canyon Located on a raised highland, This equals 0, thus serving as a hard constraint for rejection in subsequent set construction, ensuring that the final samples participating in the simulation have a completely consistent polar thermodynamic and glaciological background environment.

[0048] After completing the polar dynamics space penalty and background consistency determination, the processor filters out samples that truly conform to the laws of polar physics based on the aforementioned calculated logical equivalent distance.

[0049] Set the preset equivalent distance threshold For small units in unknown areas Iterate through the global set of known sample points, extract samples that meet the distance threshold and have the same background category, and construct an initial coarse candidate set. : Given the extreme difficulty of data acquisition in some polar regions, such as the geophysical exploration gaps deep in the Antarctic interior, strictly adhering to dual static conditions during the screening process might lead to the selection of... The internal sample size is too small to support subsequent data generalization. Therefore, the processor introduces a boundary adaptive expansion mechanism for extremely sparse polar scenarios: Preset a minimum threshold for the number of coarse screening samples. and using counting functions Calculate the total number of valid samples in the current initial coarse screening candidate set. .

[0050] like If so, it means there are a sufficient number of high-quality constraint points in the surrounding area, and the current set is output directly. As a small unit of this unknown region The final coarse candidate set.

[0051] like This indicates that the region is extremely unconstrained. In this case, the processor relaxes the distance threshold restriction, but adheres to the fundamental constraint of polar background consistency in its physical mechanism. Specifically, the processor satisfies the following across the entire region. Among the known sample points, according to the anisotropic equivalent spatial distance Sort in ascending order from smallest to largest, then forcibly truncate the first element of the sorted list. Based on the sample points, reconstruct the coarse candidate set. : in, This indicates taking the top results after sorting by equivalent distance. Operation functions for each element.

[0052] Through the aforementioned adaptive construction steps, the system tightly couples polar spatial characteristics, glacier dynamics characteristics, and geological attribute characteristics, forming a coarse-screened candidate set. This approach eliminates interference points that are physically close but cross ice shear zones or watersheds, as well as noise points with conflicting geological backgrounds. It also effectively overcomes the risk of algorithm execution interruption caused by local data gaps in polar regions. This dataset exhibits extremely high geological and logical consistency.

[0053] After completing the coarse screening of the candidate set Subsequently, this system performs residual iterative fine screening based on inversion and reconstruction of observable shallow features. In the complex evolutionary history of the polar regions, the coarse screening stage uses a candidate set defined by dynamics and background classification. Even within this model, spurious correlation samples may still be mixed in. For example, a known sample point may be highly consistent with an unknown regional unit in terms of ice flow lines and macroscopic ice facies zones, but its deeper interior lies above a local subglacial lake not yet marked on a geological base map, giving this point unique basement slip properties. This localized deep dynamic anomaly will inevitably be mapped upwards and cause changes in polar shallow surface characteristics, such as surface subsidence elevation and sudden increases in local ice fissure density. To accurately eliminate these hidden spurious correlation samples, the processor reverses the inference direction: utilizing the unknown regional unit... Observable shallow verification features already acquired through space-based remote sensing As true information, let the coarse screening set Try to reconstruct this known sample from the samples inside. If the calculation is inaccurate, it indicates the presence of bad points within the dataset that are mismatched with the local geodynamics of the current region. These need to be dynamically removed or replaced through iteration to ultimately form a high-purity, refined sample dataset. .

[0054] The specific implementation process and algorithm logic are as follows: Set the current iteration number to (initial The coarse candidate set of the current iteration is denoted as . The processor is based on the anisotropic equivalent spatial distance of the output. Calculate the known sample points within the set. For small units in unknown regions Initial reconstruction weights : in, The preset polar space attenuation index is preferably set as follows: .

[0055] Processor Extraction Set Observable shallow validation feature vectors of all sample points within the range This involves weighting and fusing column vectors containing standardized data such as elevation, slope, and albedo to calculate small units of the unknown region. In the Reconstructing shallow feature vectors in the next iteration : Extracting small units of unknown regions True observable shallow verification feature vector Calculate the reconstructed vector With the true vector Reconstruction residuals between To balance the influence of multidimensional physical quantities, the L2 norm is used to quantify vector differences: in, This represents the total number of dimensions for shallow features, such as the previously defined 5 dimensions. and The reconstructed vector and the true vector are respectively at the th... Standardized values ​​in each physical dimension. This residual. The current candidate set is comprehensively characterized. The ability to explain the physical state of this local polar environment.

[0056] To locate spurious correlated samples within the set that cause residual amplification, the processor executes a process for the current set. any known sample point The processor temporarily removes the pseudopoint from the set and recalculates the reconstruction residual after the missing pseudopoint using the remaining samples, denoted as . .

[0057] Processor defines known sample points Error contribution sensitivity : like This means that removing this point actually reduces the reconstructed residual, indicating that this point... The shallow features carried (such as anomalous fracture density or elevation) and the area to be evaluated The actual polar characteristics are contradictory, belonging to contaminated samples that are close in distance but have different physical evolutionary mechanisms.

[0058] Simultaneously, to prevent excessive loss of effective samples in polar regions, the processor introduces a suboptimal boundary sample pool. Defined as when the equivalent distance is slightly greater than a set threshold. However, edge known sample points that meet the background consistency condition are intercepted. For the sample pool... any edge point in The processor simulates adding it to the current set. Calculate the newly added reconstruction residuals after addition. The processor defines the edge points. The introduction of gain : like This indicates that the edge point is highly coordinated with the region under test in terms of deep physical mechanisms, and its introduction can neutralize the current reconstruction system error.

[0059] Based on the above decision parameters, the processor iterates through and calculates all possible values, and defines and extracts the following extreme values ​​and their corresponding extreme value sample points: Define the sensitivity of the maximum error contribution in the current iteration step. and their corresponding candidate points to be eliminated : Define the maximum introduced gain for the current iteration step. and their corresponding candidate points to be introduced : Subsequently, the processor compares the extreme parameters. and The magnitude of the value determines the dynamic update of the set. like and This indicates that removing the worst sample yields a higher benefit, so the processor performs the removal operation and updates the set. ; like and This indicates that introducing high-quality edge samples yields greater benefits, and the processor performs the introduction operation, updating the set to... At the same time, this point From the suboptimal boundary sample pool Permanently removed from [the source].

[0060] The processor iterates through the above forward reconstruction, residual comparison, sensitivity analysis, and dynamic update process (let the number of iterations be...). Until one of the following convergence conditions is triggered: Condition 1: Current reconstructed residual Less than the preset polar tolerance threshold This indicates that the sample set has perfectly fitted the local features; Condition 2: Sensitivity to maximum error contribution And the maximum introduced gain This indicates that the set state has reached Pareto optimality, and any removal or addition operation will lead to a deterioration of the fit. Condition 3: The total number of remaining samples in the set is about to fall below the minimum limit for maintaining the degrees of freedom of interpolation in polar space. (Preferred) ), forcibly terminate the removal.

[0061] When the iteration converges, the processor stops the algorithm and outputs the finally stabilized candidate set as a refined sample set. Simultaneously record the final reconstructed residual at convergence. .

[0062] Through this residual iterative screening step, the system, even when it cannot directly verify the deep properties of unknown areas, utilizes readily available shallow remote sensing data as feature probes to force a physical reconstruction test of the known sample set in the observable dimensions. The resulting refined sample set is then validated. It has been confirmed that it is related to unknown areas. A high-quality, high-value set of correlation points that share the same set of laws governing polar dynamics and thermodynamic evolution.

[0063] In polar exploration, we can discover unknown areas. The goal of regional extrapolation is to predict the deep, hidden features to be discovered. However, due to the extreme sparsity of polar data, a single predicted value often lacks decision-making reference value in engineering applications. It is necessary to simultaneously output the confidence level of the predicted value in polar space in order to quantify the uncertainty risk of the extrapolation result.

[0064] After residual iteration to remove spurious correlation points, the refined sample set Known sample points retained within Confirmed to be in an unknown area It possesses a highly coordinated polar thermodynamic evolution mechanism and glacier dynamic state. Therefore, for the deduction of concealed features, the processor constructs a static weighted deduction model based on anisotropic equivalent spatial distance.

[0065] The processor computes a fine sample set. Each known sample point within For unknown areas Final prediction weights : in, Let be the anisotropic equivalent spatial distance. This is a preset spatial attenuation coefficient for deep features. Since the smoothness of deep geological features is higher than that of surface features, The preferred value is .

[0066] Processor Extraction Set The column vector of deep hidden features to be predicted for all sample points. By combining the final prediction weights, the small units of the unknown region are obtained through weighted fusion calculation. Target hidden feature prediction column vector : in, , This indicates that the derived unit is in the first... Predicted absolute values ​​in the deep physical dimensions of the polar regions.

[0067] To scientifically evaluate the predicted column vectors of the aforementioned target concealment features To assess reliability, the processor constructed a comprehensive evaluation model for polar confidence, which includes three independent dimensions: fitting convergence, sample richness, and geological dispersion.

[0068] Fit convergence index This reflects the system's ability to reconstruct local polar features. If the system cannot accurately fit even surface features observable by space-based remote sensing, its inferences about deep ice beds are inevitably unreliable. The processor extracts the final, finalized reconstruction residual recorded at convergence. Calculate the fitting convergence index : in, This refers to the aforementioned polar tolerance threshold. When When the value approaches 0, the index approaches 1, indicating that the model achieves logical self-consistency in the observable dimension.

[0069] Sample richness index This is used to penalize the inference risks arising from the extreme scarcity of available information in polar blank areas. The processor computes a fine-grained sample set. The final number of valid samples retained Furthermore, a rich set of threshold-based saturation functions is constructed to calculate the support index. : in, To maintain an extremely low limit on the degrees of freedom for spatial interpolation (set to 3). This is the preset richness gain parameter. A high confidence level in richness is only achieved when the number of effective ice-detecting radar lines or boreholes far exceeds the baseline requirement.

[0070] Geological Dispersion Penalty Index This is used to measure the complexity of the actual subglacial geological bed surrounding an unknown region. If the depth thickness varies greatly among known points in the surrounding area (e.g., across a fault zone traversing a deeply incised subglacial canyon), the inherent risk of spatially weighted interpolation is extremely high. The processor calculates the first [unit / item] within the fine set. Local weighted variance of a hidden feature : Furthermore, to eliminate the dimensional differences between different physical quantities such as ice thickness and bottom ice temperature, the processor calculates the comprehensive variability of polar concealment features. This leads to the generation of the geological dispersion penalty index. : in, To prevent extremely small positive numbers with a denominator of zero, This is a preset polar geological dispersion tolerance penalty coefficient. The more severe the local geological variations, the higher the penalty coefficient. The larger, The data decays exponentially, thus forcibly downgrading the reliability of data in complex terrain areas.

[0071] Finally, the processor predicts column vectors of all target concealment features within the fully polar grid division. The corresponding data quality confidence scores (including fit convergence, sample richness, and geological dispersion) are packaged and transmitted to the data visualization output subsystem. The display terminal renders the numerical values ​​of deep hidden features into a basic three-dimensional contour topographic layer; at the same time, it converts the data quality confidence scores into a transparency gradient channel or a color mapping table, and synchronously overlays a confidence heatmap layer on top of the basic topographic layer.

[0072] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0073] In this specification, the same or similar parts between the various embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the descriptions of the embodiments described later are relatively simple, and relevant parts can be referred to the descriptions of the foregoing embodiments.

[0074] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A regional quality assessment method for sparse polar data, characterized in that, The method includes: Acquire polar space data and decouple the features of known sample points and unknown small units into polar space topological features, observable shallow verification features, and deep hidden features to be predicted. Based on the topological features of polar space, the anisotropic equivalent spatial distance between known sample points and unknown small units is calculated, and a coarse candidate set is constructed by combining background consistency constraints. Using classification features as hard constraints, the known sample points are calculated using a discrete variable matching function. Small units of unknown regions Background consistency constraint coefficient : ; Set the preset equivalent distance threshold Extract samples that meet the distance threshold and have a background consistency constraint coefficient of 1 to construct an initial coarse candidate set. : ； in, , These represent polar ice facies zone types and subglacial landform types, respectively. Representing small units of unknown regions The corresponding polar ice facies types and subglacial landform types; For discrete variable matching functions; The reconstruction residuals of small units in unknown regions are calculated using observable shallow verification features. The update iteration is performed based on the residual results until the convergence condition is met, and a fine sample set is generated. The update iteration based on the residual results includes the steps of eliminating spurious correlation samples and using boundary samples for collaborative gain, specifically: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] Temporarily remove a sample from the set and calculate the structural residual after the missing sample point. and define known sample points Error contribution sensitivity : ; Introducing a suboptimal boundary sample pool Suboptimal boundary sample pool For edge points, the known sample points are those whose equivalent distance is greater than a preset equivalent distance threshold and meet the background consistency condition; Simulate adding it to the current set and calculate the newly added reconstruction residual after addition. Define edge points The introduction of gain : ; The target concealment feature prediction column vector is calculated based on a fine sample set for small units in unknown regions. The data quality confidence of the target concealment feature prediction column vector is output. The data quality confidence includes three independent dimensions: fitting convergence index, sample richness index, and geological dispersion penalty index. Based on anisotropic equivalent spatial distance Calculate the fine sample set Known sample points For small units in unknown regions Final prediction weights : ; Combined with the final prediction weights For known sample points The column vector of deep hidden features to be predicted By performing weighted fusion, small units of the unknown region can be inferred. Target hidden feature prediction column vector : ; in, The preset deep feature space attenuation coefficient; The processor extracts the final reconstructed residuals recorded during convergence. Calculate the fitting convergence index : in, The set polar tolerance threshold; Sample richness index This is used to penalize the risk of inferences arising from the extreme scarcity of available information in polar blank areas: Among them, the fine sample set The total number of valid samples retained in the end is , To maintain an extremely low limit on the degrees of freedom of spatial interpolation. The preset richness gain parameter; Geological Dispersion Penalty Index Used to measure the complexity of the actual subglacial geological bed surrounding an unknown region, calculating the first [unit] within the fine set. Local weighted variance of a hidden feature : Calculate the overall variability of polar concealment features. This leads to the generation of the geological dispersion penalty index. : in, To prevent extremely small positive numbers with a denominator of zero, This is the preset polar geological discrete tolerance penalty coefficient.

2. The method according to claim 1, characterized in that, Polar space topological features Includes three-dimensional coordinate vectors Vector of surface ice flow velocity tuples: ； in, Represents known sample points Its latitude, longitude, and absolute elevation; Characterizing known sample points Two-dimensional ice flow velocity vector projected onto a horizontal plane.

3. The method according to claim 2, characterized in that, The calculation of anisotropic equivalent spatial distances between known sample points and unknown small units based on polar spatial topological features includes: By introducing the surface ice flow velocity direction vector, the physical distance is transformed into a distance that incorporates polar dynamics logic, and the known sample points are calculated. and unknown region small units Anisotropic equivalent spatial distance between : ; ; ; in, This is the streamline deviation penalty coefficient. The dynamic blocking penalty coefficient, Small units of unknown region The surface ice velocity vector at that location, This represents the highest historical ice surface flow velocity. For polar lateral stretching adjustment parameters The angle between the spatial line and the direction of the local ice flow. The spatial angle between the velocity vectors of ice flows at two points. The blocking constant of the polar watershed, The initial plane Euclidean distance is given.

4. The method according to claim 1, characterized in that, The reconstruction residuals of small units in unknown regions are calculated using observable shallow verification features, specifically including: Utilizing small units of unknown regions Using real observation data as anchor points, the current iteration step is driven. The coarse candidate set below Perform reverse physical quantity reconstruction and calculate the reconstructed shallow feature vector. : ; ; Calculate and reconstruct shallow feature vectors With real, observable, shallow verification feature vectors Reconstruction residuals between : ； in, For anisotropic equivalent spatial distance, For the initial reconstruction weights, The polar space attenuation index; For known sample points The observable shallow verification feature vector.

5. The method according to claim 4, characterized in that, The execution of the update iteration further includes the step of performing a bidirectional update of the set based on error contribution sensitivity and the introduction of gain: Extracting the sensitivity of the maximum error contribution of the current iteration step and their corresponding candidate points to be eliminated and maximum introduced gain and their corresponding candidate points to be introduced ; like and This indicates that removing spurious samples yields higher returns. Therefore, the removal operation is performed, and the candidate points to be removed are... Remove from the current collection; like and This indicates that introducing edge samples yields higher returns. Therefore, the introduction operation is performed, adding candidate points to be introduced. Add it to the current set and permanently remove it from the suboptimal boundary sample pool.

6. The method according to claim 5, characterized in that, Until the convergence condition is met, a refined sample set is generated, which specifically includes: Repeat the forward reconstruction, residual comparison, sensitivity analysis, and dynamic update process until any of the following convergence conditions are triggered. The final stable candidate set is then output as the refined sample set. Condition 1: Reconstructing the residual Less than the polar tolerance threshold Condition 2: Sensitivity to maximum error contribution and maximum introduced gain All are less than or equal to zero; Condition 3: The total number of remaining samples in the set reaches the extremely low limit for maintaining the degrees of freedom of polar space interpolation. .

7. A regional quality assessment system for polar sparse data, used to implement the regional quality assessment method for polar sparse data as described in claim 1, characterized in that, The system includes: Feature acquisition module: acquires polar space data and decouples the features of known sample points and unknown small units into polar space topological features, observable shallow verification features, and deep hidden features to be predicted; Coarse candidate set construction module: Based on the polar spatial topological features, calculate the anisotropic equivalent spatial distance between known sample points and unknown small units, and combine it with background consistency constraints to screen and construct a coarse candidate set; The fine sample set construction module calculates the reconstruction residual of the coarse candidate set for small units in the unknown region using observable shallow verification features, and performs updates and iterations based on the residual results until the convergence condition is met, thus generating the fine sample set. Evaluation module: Calculates the target hidden feature prediction column vector of small units in unknown regions based on a fine sample set, and outputs the corresponding data quality confidence score.

Images