Ozone column total amount calculation method based on air sounding observation

Abstract

The invention discloses an ozone column total amount calculation method based on sounding observation, and relates to the technical field of data analysis, by introducing a GPS three-dimensional path and air mass layer overlay analysis on the basis of traditional vertical section calculation, the situation that a sounding balloon passes through an air mass boundary or an ozone sudden change area due to transverse drift can be identified, and the overall ozone column total amount is calculated. And judging the section interfered by the transverse non-uniformity in the vertical section. By constructing a profile credibility weight mechanism, the suspected misjudgment section is flexibly corrected, the situation that transverse sudden change is misunderstood into vertical change is avoided, and the space consistency and the structure authenticity of column total amount calculation are improved. And meanwhile, a transverse interference section is extracted to generate an error sensitive mark for the remote sensing inversion model to use, so that the interference of a pseudo gradient structure on model training is remarkably reduced, and the problem of vertical misjudgment caused by balloon drift in a traditional method is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data analysis, in particular to an ozone column amount calculation method based on sounding observation. BACKGROUND

[0002] The ozone column amount calculation method based on sounding observation is usually to continuously obtain the parameters such as pressure, temperature and current response of each height layer during the ascending process of the sounding balloon carrying the ozone probe in the vertical direction, and to inversely calculate the ozone concentration corresponding to different heights according to the parameters, and finally to calculate the ozone column amount on the whole vertical path by integration. This kind of method is widely used in calibrating satellite ozone remote sensing results and supporting the research of troposphere-stratosphere atmospheric chemical process due to its strong real-time performance and high vertical resolution. The calculation process often assumes that the ozone changes gently in the horizontal direction, and the ascending path of the sounding balloon represents the real physical distribution of the vertical profile.

[0003] However, this assumption cannot always be established in the actual atmospheric environment, especially in the case of air mass intersection or ozone layer transverse structure mutation in the middle and high layers, the sounding balloon may drift horizontally and cross different concentration blocks during the ascending process, so that the "vertical variation" recorded by the instrument actually mixes the influence of "horizontal non-uniformity". This misjudgment will cause the ozone distribution which should be changed in the horizontal direction to be incorrectly mapped as a vertical structure anomaly, thereby affecting the accuracy of the column amount and the scientific interpretation of the profile form, and even may systematically interfere with the inversion accuracy of remote sensing data and subsequent climate model training. Therefore, how to identify and correct the pseudo-vertical variation introduced by horizontal drift in the sounding path has become a problem that needs to be solved but has been ignored for a long time. SUMMARY

[0004] The purpose of the present application is to solve the problems mentioned in the background art, and to propose an ozone column amount calculation method based on sounding observation.

[0005] In an embodiment of the present application, an ozone column amount calculation method based on sounding observation is provided, which comprises: S1: obtaining the GPS position corresponding to each height point in the ascending process of the sounding balloon, and constructing a three-dimensional ascending path; S2: spatially superimposing the three-dimensional ascending path and the air mass distribution layer in the meteorological reanalysis data corresponding to the time period to obtain the air mass type of each height point; S3: according to the mutation of the air mass type, identifying the height section in the ozone concentration change that may be interfered by horizontal non-uniformity, and generating a profile credibility weight; S4: introducing the profile credibility weight into the column amount calculation process, weighting and integrating the ozone concentration of each height layer to obtain the corrected ozone column amount. S5: output the lateral interference section and the corresponding weight as an error-sensitive label to support subsequent model training or remote sensing inversion error correction.

[0006] Optionally, the step of obtaining the GPS position corresponding to each height point in the ascending process of the sounding balloon to construct a three-dimensional ascending path comprises: In the sounding balloon payload system, a multi-frequency dual-channel GNSS positioning module is integrated, which is used for main signal tracking and drift compensation signal acquisition respectively, to obtain three-dimensional spatial coordinate data of the sounding balloon every second in the ascending process, including longitude, latitude and standard height of the atmosphere layer; The spatial coordinates collected every second are subjected to path trajectory smoothing processing, an adaptive Kalman filtering algorithm based on acceleration change rate constraint is adopted to identify and eliminate positioning drift abnormal points caused by sudden change of wind speed or attitude pitching, so as to obtain a three-dimensional trajectory sequence with good continuity; Based on the trajectory sequence, a three-dimensional ascending path is constructed, the standard pressure layer of each height point is bound with the corresponding GPS coordinates to form an elevation-position mapping set, and the set is subjected to spatial correction through an earth ellipsoid reference model to generate a high-precision three-dimensional ascending trajectory structure with a unified spatial reference.

[0007] Optionally, the step of spatially superimposing the three-dimensional ascending path with the air mass distribution layer in the meteorological reanalysis data of the corresponding time period to obtain the air mass type of each height point comprises: Based on the three-dimensional ascending trajectory structure, the three-dimensional coordinate information of each trajectory point in the trajectory is taken as a spatial query index, and a multi-layer air mass attribute layer in the meteorological reanalysis data set matched with the sounding time period is called; Air mass attribution identification is performed on the three-dimensional trajectory points, a layer fusion method based on voxel projection difference analysis is adopted: the three-dimensional trajectory structure is projected into the reanalysis layer coordinate system, according to the spatial angle between the trajectory point and the boundary of each grid element, the boundary tangent vector direction and the local gradient response characteristics, whether the trajectory point crosses the air mass boundary is dynamically identified, and each trajectory point is classified into the characteristic air cell currently located, and the grid domain that simultaneously satisfies the wind direction mutation, humidity jump and ozone concentration gradient inversion is preferentially selected as the boundary identification factor; The trajectory points of the air mass type are bound with the original pressure height and spatial coordinates to generate a structured air mass type label sequence.

[0008] Optionally, according to the mutation of the air mass type, the height section possibly affected by lateral non-uniform interference in the change of the ozone concentration is identified, and the step of generating a profile credibility weight comprises: Based on the ozone concentration data of adjacent height points in the three-dimensional ascending trajectory, an ozone concentration change trend index is calculated; Calculate the boundary information inconsistency index based on air mass type label sequences; Subtracting the ozone concentration change trend index from the boundary information inconsistency index yields the potential lateral error index. When the potential lateral error index exceeds a preset threshold, the corresponding height segment is identified as a potential lateral error segment. Based on the potential lateral error index corresponding to each altitude segment, profile confidence weights are generated for each altitude point in the three-dimensional ascent trajectory.

[0009] Optionally, the steps for calculating the ozone concentration change trend index based on ozone concentration data at adjacent altitude points in the three-dimensional ascent trajectory are as follows: Following the order of altitude from low to high in the three-dimensional ascent trajectory, the ozone concentration values ​​corresponding to each altitude point are obtained sequentially, and the ozone concentration difference between two adjacent altitude points is calculated to characterize the layer-by-layer change of ozone concentration in the vertical direction. For each ozone concentration difference, its direction of change is determined and converted into a trend symbol. The difference in ozone concentration that increases with altitude corresponds to a positive trend symbol, the difference in ozone concentration that decreases with altitude corresponds to a negative trend symbol, and the difference in ozone concentration that does not change corresponds to a zero trend symbol, thus forming a sequence of trend symbols arranged in order of altitude. Centered on the current altitude point, a continuous window containing several adjacent altitude points is selected in the trend symbol sequence. Within this window, the length of the longest continuous segment in which the trend symbols maintain the same direction is counted. The length of this longest continuous segment is used as a measure of the continuity of the ozone concentration change direction at the current altitude point. Within the same height range as the continuous window, the maximum and minimum values ​​of all ozone concentration differences within the window are statistically analyzed, and the difference between the two is calculated to characterize the fluctuation range of ozone concentration variation within this height range. The results of the continuity measurement of the direction of change and the fluctuation range of the amplitude of change are jointly normalized. The results of the continuity of the direction of change are proportionalized according to the window length, and the fluctuation range of the amplitude of change is normalized according to the maximum ozone concentration change amplitude in the entire three-dimensional ascent trajectory. The local trend stability value of the current altitude point is constructed by the product relationship between the two. The local trend stability values ​​corresponding to each altitude point are subjected to interval compression processing to obtain the ozone concentration change trend index.

[0010] Optionally, the steps for calculating the boundary information inconsistency index based on the air mass type label sequence are as follows: The air mass type labels are generated from the air mass type labels corresponding to each altitude point in the three-dimensional lift trajectory. The air mass type labels are classification identification codes generated after the overlay analysis of multi-source meteorological layers. For each pair of adjacent height points, iterate through them and determine whether the air mass type label of the two adjacent points has changed. If it has changed, mark the height point pair as a label abrupt change point; if it has not changed, mark it as a label continuation point. For each tagged abrupt change point, the meteorological boundary attribute information corresponding to its spatial location is extracted, including: wind direction change angle, relative humidity jump magnitude, and background ozone concentration gradient direction change. These are then evaluated. If the wind direction change exceeds 45 degrees, the humidity jump exceeds 15%, and the ozone concentration gradient direction reverses, the boundary consistency state value for that tagged abrupt change point is assigned to 1. If only two of the above three conditions are met, the state value is 0.66; if only one condition is met, the state value is 0.33; if none of these conditions are met, the state value is 0. The sum of the boundary consistency values ​​of all labeled mutation points is calculated and divided by the number of labeled mutation points to obtain the average boundary consistency value; the number of labeled mutation points is divided by the total number of adjacent height points to obtain the label mutation ratio value; the two values ​​are multiplied together to obtain the boundary mutation significance. Further, count the number of points among all tag continuation points where the air mass tag remains consistent but the absolute value of the ozone concentration difference is greater than the 95th percentile of the total trajectory concentration difference sequence, and divide this number by the total number of tag continuation points to obtain the structural drift mismatch degree; The boundary information inconsistency index is obtained by adding the significance of boundary mutations to the structural drift mismatch and then multiplying by one-half.

[0011] Optionally, the steps to incorporate profile confidence weights into the total ozone column calculation process, and to obtain the corrected total ozone column volume by weighted integration of ozone concentrations at each height level, are as follows: Based on the profile confidence weight sequence generated at each altitude point, each altitude point is divided into several continuous integration units according to its order in the ascent trajectory. Each integration unit includes a set of altitude points and their corresponding ozone concentration values ​​and profile confidence weight values. For each integration unit, determine whether the profile confidence weight of all height points inside it is greater than the preset full confidence threshold. If so, retain the complete height range of the integration unit and multiply its corresponding ozone concentration value by the corresponding physical thickness as a standard integration term to participate in the total column calculation. If there is a point in the integration unit where the profile confidence weight is lower than the preset local anomaly threshold, the system will activate the integration back-off mechanism: that is, expand a fixed height window upward and downward near the anomaly point. If the confidence in this interval shows a continuous downward trend, the data segment will be removed and will not participate in the integration. If there is an anomaly in the middle or confidence at both ends, only the confidence segments at both ends will be retained and the integration interval will be compressed according to the confidence weight. For the retained integral units, a weight control function is introduced instead of direct weighting. The control function maps the profile confidence weight value to an integral coefficient adjustment factor. The adjustment method is as follows: when the weight value is greater than the high confidence threshold, the integral coefficient remains unchanged; when the weight value is between the high confidence and the abnormal threshold, the integral coefficient decays exponentially; when the weight value is lower than the abnormal threshold, the integral coefficient is set to zero, and this height point is not included in the total column. The sum of the integral results after eliminating, compressing and adjusting all integral units is used as the corrected total ozone column value.

[0012] Optionally, the steps for outputting the lateral interference segments and their corresponding weights as error-sensitive labels to support subsequent model training or remote sensing inversion error correction are as follows: Based on the identified potential lateral error segments and the corresponding profile confidence weight sequence, the altitude points in the three-dimensional launch trajectory are reconstructed in segments. The intervals in continuous altitude points with profile confidence weights lower than the preset labeling threshold are merged into lateral interference segments, and a unique segment identifier is generated for each lateral interference segment. For each lateral interference segment, the starting and ending heights of the segment, the minimum and average values ​​of the profile confidence weights within the segment, and the gradient features of weight changes are extracted. This information is then encapsulated into structured error annotation units. The error annotation unit is bound to the corresponding three-dimensional launch trajectory coordinates, sounding timestamp, and ozone column total calculation results to generate a traceable error-sensitive annotation record, so that each annotation segment can be back located to the specific height range in the original sounding path. Error-sensitive annotations are mapped to the model training interface and the remote sensing inversion correction interface, respectively. During model training, error-sensitive annotations are used as sample confidence constraint information to reduce the influence weight of data in the corresponding height segment in model parameter updates. During remote sensing inversion correction, error-sensitive annotations are used as profile confidence masks to limit the constraint strength of the inversion algorithm in the corresponding height segment. The output is a set of error-sensitive labels containing the location of the lateral interference segment, the corresponding profile confidence weight, and its labeling attributes, which serves as unified input data for subsequent model training and remote sensing inversion error correction.

[0013] The beneficial effects of this invention are: This invention proposes a method for calculating total ozone column volume based on radiosonde observations. By introducing GPS 3D path recordings and air mass layer overlay analysis on top of traditional vertical profile calculations, it can accurately identify air mass boundaries or ozone concentration abrupt change zones that radiosonde balloons may traverse during ascent due to lateral drift. This allows for the determination of which vertical altitude segments are affected by lateral non-uniformity in ozone concentration changes. Furthermore, by constructing a profile reliability weighting mechanism, flexible corrections are implemented for potentially misjudged segments. This ensures that the total volume integration process retains the continuity of the original data while avoiding misinterpreting lateral abrupt changes as vertical structural changes, effectively improving the spatial consistency and structural accuracy of ozone column volume estimation. In addition, this method outputs lateral interference information as error-sensitive labels, providing reliable constraints for remote sensing inversion model training and significantly reducing the interference of pseudo-gradient structures on model fitting ability. This fundamentally solves the long-standing implicit problem in the background technology that "misjudgment of vertical changes originates from lateral drift." Attached Figure Description

[0014] Figure 1 A flowchart illustrating the method for calculating total ozone column volume based on radiosonde observations provided in this embodiment of the invention. Detailed Implementation

[0015] To further illustrate the technical means and effects adopted by the present invention in order to achieve the intended purpose, the following detailed description is provided in conjunction with the accompanying drawings and preferred embodiments, based on the specific implementation methods, structures, features and effects of the present invention.

[0016] The present invention provides a method for calculating total ozone column concentration based on radiosonde observations. See also... Figure 1 , Figure 1 A flowchart illustrating a method for calculating total ozone column density based on radiosonde observations, provided in an embodiment of the present invention. The method includes the following steps: S1: Obtain the GPS positions at various altitudes during the launch of the weather balloon and construct a three-dimensional launch path; S2: Spatially overlay the three-dimensional ascent path with the air mass distribution layer in the meteorological reanalysis data of the corresponding time period to obtain the air mass type at each altitude point; S3: Based on the abrupt changes in air mass type, identify the height segments in ozone concentration changes that may be affected by lateral non-uniformity and generate profile confidence weights. S4: Introduce the profile confidence weight into the total column calculation process, perform weighted integration of ozone concentration at each height layer, and obtain the corrected total ozone column. S5: Output the lateral interference segment and its corresponding weight as an error-sensitive label to support subsequent model training or remote sensing inversion error correction.

[0017] Based on the ozone column total calculation method based on radiosonde observation provided in this invention, by introducing GPS 3D path recording and air mass layer overlay analysis on the basis of traditional vertical profile calculation, it can accurately identify the air mass boundaries or ozone concentration abrupt change areas that the radiosonde may cross due to lateral drift during ascent, thereby determining which vertical height segments are affected by lateral non-uniformity in ozone concentration changes. Furthermore, by constructing a profile credibility weighting mechanism, flexible corrections are implemented for segments that may be misjudged, so that the total integration process can avoid mistaking lateral abrupt changes as vertical structural changes while preserving the continuity of the original data, effectively improving the spatial consistency and structural authenticity of ozone column total estimation. In addition, this method also outputs lateral interference information as error-sensitive labels, providing reliable constraints for remote sensing inversion model training, significantly reducing the interference of pseudo-gradient structures on model fitting ability, and fundamentally solving the long-standing implicit problem in the background technology that "misjudgment of vertical changes originates from lateral drift".

[0018] In one embodiment, S1: The steps for obtaining the GPS positions corresponding to each altitude point during the launch of the weather balloon and constructing the three-dimensional launch path are as follows: A multi-band dual-channel GNSS positioning module is integrated into the sounding balloon payload system, which is used for main signal tracking and drift compensation signal acquisition, respectively, to obtain three-dimensional spatial coordinate data of the sounding balloon every second during the ascent process. The three-dimensional spatial coordinates include longitude, latitude and standard atmospheric altitude. The spatial coordinates collected every second are processed to smooth the path trajectory. An adaptive Kalman filter algorithm based on the acceleration rate of change constraint is used to identify and remove abnormal positioning drift points caused by sudden changes in wind speed or attitude turbulence, so as to obtain a three-dimensional trajectory sequence with good continuity. A three-dimensional ascent path is constructed based on the trajectory sequence. The standard pressure layer at each altitude point is bound to its corresponding GPS coordinates to form an elevation-position mapping set. The set is then spatially corrected using an Earth ellipsoid reference model. Finally, a high-precision three-dimensional ascent trajectory structure with a unified spatial benchmark is generated for subsequent air mass matching and disturbance identification.

[0019] It should be noted that the process of acquiring the GPS positions at various altitudes during the launch of the weather balloon and constructing a three-dimensional launch path employs an innovative approach with spatial drift compensation and high-precision trajectory construction capabilities. First, a multi-band dual-channel GNSS positioning module is integrated into the weather balloon's payload system. One channel continuously and stably tracks the primary positioning signal, while the other independently acquires auxiliary signals to capture short-term drift trends. This effectively suppresses positioning errors caused by complex ionospheric conditions in the upper atmosphere or changes in balloon attitude, ensuring that the positioning results have second-level resolution and high spatial consistency. The resulting raw three-dimensional coordinate data includes longitude, latitude, and atmospheric standard altitude (such as geographic elevation converted from MSL), and a preliminary spatial point series is formed with a sampling period of one second. Next, an adaptive Kalman filter algorithm with fused acceleration rate of change constraints is used to smooth the path trajectory of this spatial point series. This algorithm not only considers conventional speed and azimuth continuity but also actively identifies spatial jumps caused by wind shear, turbulence, or balloon rotation, such as the positioning jumps that often occur when crossing rapidly changing pressure layers or strong wind layers. Such outliers are difficult to remove using conventional filtering, easily leading to spurious vertical transitions. This algorithm, however, introduces the rate of change of acceleration as a dynamic constraint factor, dynamically adjusting the filter gain to effectively suppress spurious disturbances and improve trajectory continuity. The filtered trajectory sequence is further bound to the standard pressure layer corresponding to each moment, constructing an elevation-position mapping set. Based on this, a geoid reference model (such as WGS84) is used to perform spatial projection correction on the trajectory, resolving the spatial offset problem caused by different coordinate references. Finally, a complete, coordinate-uniform, and geometrically accurate three-dimensional ascent trajectory is generated, which can be used for subsequent spatial overlay with reanalysis air mass layers and to identify whether the trajectory crosses a lateral boundary of ozone concentration abrupt change in the vertical path. For example, in a sounding mission, if a balloon drifts horizontally at an altitude of 12 kilometers and simultaneously crosses the boundary between a tropical maritime air mass and a high-latitude dry and cold air mass, traditional methods would misinterpret it as a sudden change in vertical ozone concentration. However, the three-dimensional trajectory combined with the drift point identification mechanism described in this invention will correctly identify the change as being caused by lateral crossing, thus providing a highly reliable spatial basis for subsequent error sensitivity analysis.

[0020] In one embodiment, S2: The step of spatially overlaying the three-dimensional ascent path with the air mass distribution layer in the meteorological reanalysis data of the corresponding time period to obtain the air mass type at each altitude point is as follows: Based on the three-dimensional launch trajectory structure with a unified spatial reference constructed in step S1, the three-dimensional coordinate information of each trajectory point in the trajectory is used as a spatial query index to call the multi-layer air mass attribute layer in the meteorological reanalysis dataset that matches the sounding time period. The air mass attribute layer includes the wind field zoning map under the isobaric surface, the humidity block boundary map, the geopotential height field, and the ozone concentration reanalysis grid data. To identify the air mass affiliation of 3D trajectory points, a layer fusion method based on "voxel projection difference analysis" is adopted: the 3D trajectory structure is projected onto the coordinate system of the reanalysis layer, and the trajectory points are dynamically identified as crossing the air mass boundary according to the spatial angle between the trajectory points and the boundaries of each grid cell, the direction of the boundary tangent vector, and the local gradient response characteristics. Each trajectory point is then classified into the characteristic air mass cell in which it is currently located, and grid domains that simultaneously satisfy the conditions of sudden wind direction change, sudden humidity change, and ozone concentration gradient reversal are preferentially selected as boundary identification factors. By binding the trajectory points that have been assigned to air mass types with their original pressure altitude and spatial coordinates, a structured air mass type label sequence is generated. This sequence corresponds one-to-one with the three-dimensional ascent trajectory structure and serves as the basic input for subsequent identification of lateral non-uniform interference sections.

[0021] It should be noted that the process of spatially overlaying the three-dimensional launch path with the air mass distribution layer in the meteorological reanalysis data of the corresponding time period to obtain the air mass type at each altitude point adopts a combination of multi-source layer fusion and spatial difference projection, achieving high-resolution identification of air mass affiliation. First, based on the three-dimensional launch trajectory structure with a unified spatial benchmark constructed in step S1, the longitude, latitude, and standard atmospheric altitude triplet of each altitude point in the trajectory are used as spatial query indexes. According to the time matching principle, the corresponding multi-layer air mass attribute layers in the global meteorological reanalysis database (such as ERA5 or MERRA2) are synchronously called. These layers include, but are not limited to, isobaric wind field zoning maps (used to identify large-scale air mass dynamic boundaries), humidity block boundary maps (used to capture differences in water vapor structure), geopotential height fields (to assist in judging the hierarchical structure of air masses), and ozone concentration reanalysis grid data (as a reference for sounding data comparison). Next, a method based on "voxel projection difference analysis" is used to determine the air mass affiliation. This method does not simply insert trajectory points into the grid to obtain values; instead, it projects the three-dimensional ascent trajectory structure onto the three-dimensional grid voxel coordinate system of the reanalysis layer and calculates the spatial angle between the trajectory point and the adjacent grid boundary, the orientation of the boundary normal / tangent vector, and the gradient change magnitude of local parameters to determine whether the current trajectory point is located at the air mass boundary or within the air mass. If the wind direction corresponding to a trajectory point changes by more than 45 degrees between two adjacent grids, the relative humidity jumps by more than 15%, or the ozone concentration gradient reverses (from positive to negative or vice versa), it is determined that it crosses the air mass boundary, and the trajectory point is classified into a new air mass type unit accordingly; otherwise, the trajectory point is considered to belong to the original air mass continuous interval. This not only avoids misclassification due to interpolation ambiguity but also sensitively identifies microscale abrupt changes in structure below the air mass boundary. For example, when a weather balloon ascends to approximately 11.5 km above the tropopause, if a sudden change in wind direction is accompanied by a sharp drop in humidity, this point will be identified as the "tropospheric-stratospheric transitional air mass boundary," thus distinguishing it from its upper and lower layers. Ultimately, the air mass type of each trajectory point is bound to its corresponding pressure altitude and geographic coordinates, constructing a complete air mass type label sequence. This achieves a one-to-one mapping between the three-dimensional trajectory and air mass distribution, providing accurate spatial basis and semantic identifiers for subsequent lateral non-uniformity identification and profile correction. Compared to traditional grid interpolation-based classification methods, this method is more suitable for identifying the attribution of regions with drastic boundary changes, frontal regions, and multi-field coupled structures such as ozone fronts, demonstrating significant spatial resolution and dynamic discrimination capabilities.

[0022] In one embodiment, S3: The step of identifying height segments in ozone concentration changes that may be affected by lateral non-uniformity based on abrupt changes in air mass type and generating profile confidence weights includes: Based on ozone concentration data at adjacent altitudes in the three-dimensional ascent trajectory, an ozone concentration change trend index is calculated. The ozone concentration change trend index is used to characterize the directional consistency and stability of the change amplitude of ozone concentration between adjacent altitude layers, so as to reflect the continuous evolution characteristics of ozone in the vertical direction. Based on the air mass type label sequence obtained in step S2, the boundary information inconsistency index is calculated. The boundary information inconsistency index is used to characterize whether there is a sudden change in air mass type between adjacent height points, and whether the sudden change corresponds to the disruption of the continuity of air mass boundary properties. Subtracting the ozone concentration change trend index from the boundary information inconsistency index yields the potential lateral error index, which is used to quantify the degree of consistency deviation between ozone concentration changes and air mass structure changes. Based on the magnitude of the potential lateral error index, it is determined whether each height point belongs to a height segment that may be affected by lateral non-uniformity. When the potential lateral error index exceeds a preset threshold, the corresponding height segment is identified as a potential lateral error segment. Based on the potential lateral error index corresponding to each altitude segment, a profile confidence weight is generated for each altitude point in the three-dimensional ascent trajectory. The profile confidence weight is used to characterize the reliability of the ozone concentration data at that altitude point in the subsequent total column calculation.

[0023] In one implementation, the steps for calculating the ozone concentration change trend index based on ozone concentration data at adjacent altitude points in the three-dimensional ascent trajectory are as follows: Following the order of altitude from low to high in the three-dimensional ascent trajectory, the ozone concentration values ​​corresponding to each altitude point are obtained sequentially, and the ozone concentration difference between two adjacent altitude points is calculated to characterize the layer-by-layer change of ozone concentration in the vertical direction. For each ozone concentration difference, its direction of change is determined and converted into a trend symbol. The difference in ozone concentration that increases with altitude corresponds to a positive trend symbol, the difference in ozone concentration that decreases with altitude corresponds to a negative trend symbol, and the difference in ozone concentration that does not change corresponds to a zero trend symbol, thus forming a sequence of trend symbols arranged in order of altitude. Centered on the current altitude point, a continuous window containing several adjacent altitude points is selected in the trend symbol sequence. Within this window, the length of the longest continuous segment in which the trend symbols maintain the same direction is counted. The length of this longest continuous segment is used as a measure of the continuity of the ozone concentration change direction at the current altitude point. Within the same height range as the continuous window, the maximum and minimum values ​​of all ozone concentration differences within the window are statistically analyzed, and the difference between the two is calculated to characterize the fluctuation range of ozone concentration variation within this height range. The continuity measurement result of the change direction obtained in step 3 and the fluctuation range of the change amplitude obtained in step 4 are jointly normalized. The continuity result of the change direction is proportionalized according to the window length, and the fluctuation range of the change amplitude is normalized according to the maximum ozone concentration change amplitude in the entire three-dimensional ascent trajectory. The local trend stability value of the current altitude point is constructed by the product relationship between the two. The local trend stability values ​​corresponding to each altitude point are subjected to interval compression processing to obtain the ozone concentration change trend index in the range of 0 to 1. The closer the value of the ozone concentration change trend index is to 1, the stronger the consistency of the ozone concentration change direction with altitude at that altitude point and the smoother and more continuous the change process. The closer the value is to 0, the more frequently the ozone concentration change direction reverses or the change amplitude fluctuates violently at that altitude point, and the vertical change trend is unstable.

[0024] It should be noted that in the calculation of the ozone concentration change trend index mentioned above, the data involved in each calculation step are all based on the original observation records of the sounding balloon during its ascent and their corresponding altitude sequences, with clear and verifiable sources. Specifically, the ozone concentration data first comes from the electrochemical ozone probe carried by the sounding balloon. This probe records the atmospheric ozone concentration value at each altitude point along the ascent path at a fixed sampling frequency (e.g., once per second). The raw data is received in real time by the sounding ground receiving system and paired one-to-one with barometric altitude or GPS altitude to form a complete "altitude-ozone concentration" sequence. Altitude data can come from standard barometric altitude calculation (using atmospheric pressure measured by a barometric sensor and converted to altitude using an international standard atmospheric model) or geometric altitude provided by a GNSS module, and is unified to the same reference system as needed. Based on this, in order to determine the trend direction and calculate the difference, the system calculates the difference in ozone concentration between each pair of adjacent altitude points in ascending order of altitude; then, the direction of change (increasing, decreasing, or unchanged) corresponding to each difference is marked as a positive, negative, or zero trend sign. To obtain a continuity index for local trends, the system selects a sliding window of fixed length (e.g., 5 or 7 points) from the trend symbol sequence, centered on a target altitude point. It analyzes whether the trend symbols of adjacent points within the window remain consistent, and calculates the length of the longest continuous segment with consistent direction as the directional continuity measure. The difference data used within this window comes directly from the aforementioned continuous difference sequence. Subsequently, to obtain the fluctuation intensity, the system analyzes the maximum and minimum values ​​of all differences within the same window and calculates the difference as the concentration fluctuation amplitude. This amplitude is then normalized using the maximum absolute value of all differences throughout the entire ascent trajectory. These normalized reference values ​​are calculated from the differences of all sampling points during ascent, without requiring external unknown factors. Finally, the above trend continuity measure is normalized by the window length (e.g., divided by the window length), multiplied by the normalized fluctuation amplitude result to obtain the local trend stability, which is directly used as the ozone concentration change trend index at the current altitude point. The entire data collection, processing, and indexing process is completed continuously within a fixed time series, without relying on subjective judgment or introducing weighting factors or external assumptions, thus ensuring the physical interpretability of the index and the engineering feasibility of the calculation process. For example, if in a balloon launch record, the concentration in the 8.5km to 9.2km altitude range continuously decreases with a consistent trend sign and a fluctuation range less than 20% of the overall maximum fluctuation, the system can then provide a highly reliable output with a trend index close to 1 for that altitude range.

[0025] It should be noted that the ozone concentration change trend index is an indicator used to measure whether the ozone concentration changes at various altitudes during the vertical ascent of a weather balloon exhibit "directional consistency" and "amplitude stability." Essentially, it reflects whether the continuous evolution of ozone concentration across a vertical profile is smooth and follows the expected vertical diffusion or interlayer gradient. A large ozone concentration change trend index at a given altitude indicates that the ozone concentration change direction at that point and between adjacent altitudes is consistent (e.g., continuous increase or decrease), and the amplitude of change is relatively stable without drastic jumps. This signifies good vertical trend continuity in the profile and is more likely to represent the true vertical distribution process, with a lower probability of being affected by lateral non-uniformity. Conversely, a small trend index indicates frequent reversals in the concentration change direction near that point, alternating increases and decreases, or drastic fluctuations in the concentration amplitude. This may not be a natural result of vertical evolution but rather a result of the balloon laterally crossing air mass boundaries or local abrupt change zones, causing abrupt ozone concentration changes and forming a "pseudo-gradient." In this case, the risk of lateral non-uniformity interference at that altitude point increases significantly. For example, during a sounding event, if the ozone concentration at an altitude of 9.3 km shows a continuous increase, but then suddenly decreases at that point and increases again at the next point, reversing the trend direction and exhibiting a much larger fluctuation amplitude than other segments, the trend index at that point will decrease significantly. This suggests that the concentration at that point may not be a true vertical abrupt change, but rather a result of lateral drift into another air mass. This pattern of decreasing trend index is a crucial signal for identifying lateral non-uniform disturbance segments. Therefore, the smaller the ozone concentration change trend index, the more unstable the vertical concentration change at that altitude point is, and the more likely it is to be disturbed by lateral air mass structural abrupt changes. A lower profile reliability weight should be assigned to this index. This index not only helps identify anomalous disturbance points but also serves as an important basis for the integral weight of the total control column, improving the spatial consistency and reliability of the overall profile data from the source.

[0026] In one implementation, the step of calculating the boundary information inconsistency index based on the air mass type label sequence obtained in step S2 is as follows: In step S2, obtain the air mass type label corresponding to each altitude point in the three-dimensional ascent trajectory. The air mass type label is a classification identifier code generated after the multi-source meteorological layer overlay analysis. For each pair of adjacent height points, iterate through them and determine whether the air mass type label of the two adjacent points has changed. If it has changed, mark the height point pair as a label abrupt change point; if it has not changed, mark it as a label continuation point. For each tagged abrupt change point, the meteorological boundary attribute information corresponding to its spatial location is extracted, including: wind direction change angle, relative humidity jump magnitude, and background ozone concentration gradient direction change. These are then evaluated. If the wind direction change exceeds 45 degrees, the humidity jump exceeds 15%, and the ozone concentration gradient direction reverses, the boundary consistency state value for that tagged abrupt change point is assigned to 1. If only two of the above three conditions are met, the state value is 0.66; if only one condition is met, the state value is 0.33; if none of these conditions are met, the state value is 0. The sum of the boundary consistency values ​​of all labeled mutation points is calculated and divided by the number of labeled mutation points to obtain the average boundary consistency value; the number of labeled mutation points is divided by the total number of adjacent height points to obtain the label mutation ratio value; the two values ​​are multiplied together to obtain the boundary mutation significance. Further, count the number of points among all tag continuation points where the air mass tag remains consistent but the absolute value of the ozone concentration difference is greater than the 95th percentile of the total trajectory concentration difference sequence, and divide this number by the total number of tag continuation points to obtain the structural drift mismatch degree; The boundary information inconsistency index is obtained by adding the boundary abrupt change significance to the structural drift mismatch and multiplying by one-half. The boundary information inconsistency index is a dimensionless quantity between 0 and 1. The larger the value, the weaker the consistency between the change in air mass label and the change in ozone concentration, and the greater the possibility that the corresponding height point in the vertical profile is affected by lateral non-uniformity.

[0027] In the calculation of the boundary information inconsistency index, all data involved are obtained by structurally fusing real-time observation data collected during the launch of the sounding balloon with meteorological reanalysis products, possessing clear data sources and acquisition paths. First, the air mass type label at each altitude point in the three-dimensional launch trajectory is obtained by overlaying the trajectory points from step S2 with the meteorological reanalysis data space. Specifically, this involves using the GPS three-dimensional spatial coordinates obtained per second by the sounding balloon as an index, matching the wind field zoning map, relative humidity field, geopotential height field, and ozone concentration grid at the corresponding time in ERA5, MERRA2, or similar high spatiotemporal resolution reanalysis datasets, and determining the regional affiliation of the trajectory points based on the wind direction axis, humidity mass boundaries, and concentration classification boundaries, thereby generating air mass classification labels, such as air mass A, air mass B, etc., using a coding method. After the trajectory label sequence is constructed, the system compares the labels of adjacent altitude points one by one to determine whether label changes have occurred, and classifies them as "label mutation points" or "label continuation points" accordingly. For each tagged abrupt change point, the system further extracts the wind direction vector difference, relative humidity difference, and ozone concentration gradient change direction (e.g., from positive to negative gradient) of the spatial location of that point in the reanalysis data. These three types of information are calculated from the vector wind field data, humidity profile field, and ozone concentration grid, respectively, without relying on external assumptions or estimation processes. The system compares these three actual physical variables with a preset jump threshold and assigns a boundary consistency state value to reflect the significance of the structural abrupt change. Afterward, the system statistically analyzes the mean and proportion of state values ​​for all tagged abrupt change points, which can be directly deduced from the trajectory point label sequence and the total number of samples without inference. Furthermore, to further identify regions with "consistent surface structure but actual concentration abrupt changes," the system extracts the corresponding ozone concentration data (from the electrochemical ozone probe of a weather balloon) from all tagged continuation points, calculates the difference between adjacent points, and compares it with the 95th percentile of the entire trajectory difference sequence to determine whether it belongs to an "abnormal concentration jump." This quantile value is obtained statistically from the trajectory's own data without external intervention. For example, if the concentration difference at a set of label continuation points exceeds the variation range at most locations along the entire trajectory, it is identified as a potential mismatch structure region. Finally, the system normalizes and integrates the aforementioned "boundary abrupt change significance" and "structural drift mismatch degree" according to a fixed formula to form the final boundary information inconsistency index, used to determine the degree of consistency between the profile structure and concentration evolution. All data acquisition and judgment are completed between the radiosonde trajectory and the standardized meteorological reanalysis database, possessing complete physical basis and engineering feasibility.

[0028] The Boundary Information Inconsistency Index (BII) is a dimensionless indicator used to measure the inconsistency between changes in air mass structure and ozone concentration at adjacent altitudes on a vertical profile during weather balloon ascent. Its essential purpose is to determine whether vertical ozone changes at a given altitude may be affected by lateral non-uniform structures (such as air mass boundaries or frontal overlap). Specifically, the index is constructed by considering two scenarios: first, whether abrupt changes in air mass type labeling are accompanied by typical changes in boundary attributes (such as large-angle changes in wind direction, abrupt changes in humidity, or reversals in background ozone concentration); and second, situations where the air mass label remains unchanged but ozone concentration changes drastically, indicating that this structural drift mismatch may be a profile distortion caused by lateral drift. A higher BII value indicates more significant inconsistencies in both scenarios, suggesting that the change at that altitude may not represent a true vertical concentration evolution but rather is influenced by lateral structural disturbances. Conversely, a lower BII value indicates a stable and consistent relationship between the air mass label and concentration change trends, suggesting high reliability. For example, in a certain profile segment, if the air mass label frequently jumps from an altitude of 8.2 km to 8.6 km, but the corresponding ozone concentration does not change significantly, or conversely, if the concentration jumps abruptly but the label does not change, the boundary information inconsistency index will increase significantly. This indicates that the structural behavior of this segment may originate from the horizontal displacement of the weather balloon rather than a true change in the vertical layers, thus requiring a lower profile confidence weight to be assigned to this segment. Therefore, in practical applications, the larger the boundary information inconsistency index, the greater the likelihood that the corresponding altitude point belongs to an altitude segment that may be subject to lateral non-uniform interference. This index can serve as an important basis for dynamically adjusting the weights in the column total integral and also provides key structural discrimination information for subsequent remote sensing data comparison and model training.

[0029] In one embodiment, S4: The step of incorporating profile confidence weights into the total ozone column calculation process and performing a weighted integral of the ozone concentration at each height level to obtain the corrected total ozone column is as follows: Based on the profile confidence weight sequence generated for each altitude point in step S3, each altitude point is divided into several continuous integration units according to its order in the ascent trajectory. Each integration unit includes a set of altitude points and their corresponding ozone concentration values ​​and profile confidence weight values. For each integration unit, determine whether the profile confidence weight of all height points inside it is greater than the preset full confidence threshold. If so, retain the complete height range of the integration unit and multiply its corresponding ozone concentration value by the corresponding physical thickness as a standard integration term to participate in the total column calculation. If there is a point in the integration unit where the profile confidence weight is lower than the preset local anomaly threshold, the system will activate the integration backtracking mechanism: that is, expand a fixed height window (e.g. ±100 meters) upwards and downwards near the anomaly point, evaluate the confidence gradient trend in the interval, if the confidence shows a continuous downward trend, then remove the data segment and do not participate in the integration, if it shows a V-shaped or U-shaped structure (i.e., anomaly in the middle and confidence at both ends), then only the confidence segments at both ends are retained and the integration interval is compressed according to the confidence weight; For the retained integral units, a weight control function is introduced instead of direct weighting. The control function maps the profile confidence weight value to an integral coefficient adjustment factor. The adjustment method is as follows: when the weight value is greater than the high confidence threshold, the integral coefficient remains unchanged; when the weight value is between the high confidence and the abnormal threshold, the integral coefficient decays exponentially; when the weight value is lower than the abnormal threshold, the integral coefficient is set to zero, and this height point is not included in the total column. The integration results of all integration units after elimination, compression and adjustment are summed up to obtain the corrected total ozone column value. The total column value has adaptive fault tolerance capability for cross-sectional interference points, avoiding structural deviations caused by single-point errors in the total calculation, while maintaining the physical consistency and integration continuity of the total column value.

[0030] It should be noted that the key to introducing profile confidence weights into the total column calculation process lies in effectively avoiding and controlling local profile anomalies caused by lateral non-uniformity interference by dynamically adjusting the contribution value of each height point in the integration path to the total column. First, based on the profile confidence weight sequence calculated in step S3, the system divides each height point in the three-dimensional ascent trajectory according to the ascent order, constructing several continuous integration units. Each integration unit consists of multiple adjacent height points, and records the ozone concentration value, physical thickness, and corresponding profile confidence weight for each point. For each integration unit, the system first determines whether the profile confidence weights of all points within it are higher than the set full confidence threshold (e.g., 0.9). If so, the unit is treated as a high-confidence segment, retains its complete structure, and directly participates in the total column calculation, using the standard concentration multiplied by layer thickness integration method. When the weight of any point within the integration unit falls below a set local anomaly threshold (e.g., 0.5), the system triggers an integration foldback mechanism: It extends a fixed height range (e.g., ±100 meters) upwards and downwards from the anomaly point and analyzes the gradient trend of profile confidence within this local range. If a continuous decrease in confidence is detected, it is considered a profile disturbance area, and this segment of data is completely removed and not included in the integration. If the segment exhibits a V-shaped or U-shaped structure (i.e., anomaly in the middle but good recovery of weights on both sides), the system retains only the two reliable boundary segments and compresses the integration layer thickness of this segment to reduce the impact of errors. For the retained integration data points, instead of directly weighting them according to the original weights, a weight control function is introduced to map profile confidence to an integration adjustment factor: when the weight value of a certain height point is higher than the high confidence threshold, the concentration value of that point multiplied by its layer thickness is retained as a normal integration term; when the weight is in the intermediate range between the high confidence and the anomaly threshold, the system compresses the integration contribution of that point according to an exponential decay function; when the weight is lower than the anomaly threshold, the integration contribution of that point is directly set to zero, equivalent to complete masking. Through this control function, the system achieves adaptive regulation of local errors in the total column integration. This avoids both the systematic errors caused by the insensitivity of weights in traditional weighted integration methods and the structural discontinuities caused by simple elimination. For example, when the balloon ascends to 11.7 km and encounters a significant abrupt change in concentration due to a lateral air mass mutation, and the profile confidence weight is less than 0.3, the system will automatically activate a backtracking mechanism to eliminate this segment and perform local compression processing on the adjacent confidence region to ensure that the total column result is not amplified by local distortion interference. Finally, the results of all processed integration units are accumulated to form a corrected total ozone column value with structural repair and fault tolerance capabilities, effectively improving the computational reliability of the entire profile in the presence of lateral non-uniformity.

[0031] In one embodiment, S5: The step of outputting the lateral interference segment and its corresponding weight as an error-sensitive label to support subsequent model training or remote sensing inversion error correction is as follows: Based on the potential lateral error segments and corresponding profile confidence weight sequences identified in step S3, the altitude points in the three-dimensional launch trajectory are reconstructed in segments. Intervals with profile confidence weights lower than the preset labeling threshold in continuous altitude points are merged into lateral interference segments, and a unique segment identifier is generated for each lateral interference segment. For each transverse interference segment, the starting and ending heights of the segment, the minimum and average values ​​of the profile confidence weights within the segment, and the gradient features of weight changes are extracted. The above information is then encapsulated into structured error annotation units, which are used to fully describe the spatial range and interference intensity characteristics of the segment. The error annotation unit is bound to the corresponding three-dimensional launch trajectory coordinates, sounding timestamp, and ozone column total calculation results to generate a traceable error-sensitive annotation record, so that each annotation segment can be back located to the specific height range in the original sounding path. Error-sensitive annotations are mapped to the model training interface and the remote sensing inversion correction interface, respectively. During model training, error-sensitive annotations are used as sample confidence constraint information to reduce the influence weight of data in the corresponding height segment in model parameter updates. During remote sensing inversion correction, error-sensitive annotations are used as profile confidence masks to limit the constraint strength of the inversion algorithm in the corresponding height segment. The output is a set of error-sensitive labels containing the location of the lateral interference segment, the corresponding profile confidence weight, and its labeling attributes, which serves as unified input data for subsequent model training and remote sensing inversion error correction.

[0032] It should be noted that the process of outputting the lateral interference segments and their corresponding profile confidence weights as error-sensitive labels mainly aims to structurally identify and embed the height segments in the weather balloon profile affected by lateral non-uniform interference into the downstream application system, thereby achieving local suppression of remote sensing inversion errors and confidence constraint control during model training. Specifically, the system first locates multiple continuous height point intervals in the profile with weights significantly lower than a preset labeling threshold (e.g., 0.4) based on the potential lateral error index results identified in step S3. Each such continuous segment is considered a lateral interference segment and assigned a unique number for subsequent tracking and binding. Next, for each lateral interference segment, the system extracts its starting height, ending height, the minimum and average values ​​of all profile confidence weights within the segment, and their vertical slope (i.e., gradient). This information is then integrated into a structured "error labeling unit," which not only represents spatial location but also the intensity of interference, possessing rich information dimensions. Subsequently, the system binds each error annotation unit to its three-dimensional spatial coordinates (i.e., the GPS trajectory of the radiosonde in that section), the timestamp of the radiosonde mission, and the final output column total value, constructing a complete "error-sensitive annotation record" to achieve traceability and reverse positioning. To improve the practicality of the annotation results, this record is designed to be bidirectionally embedded into downstream processes: in remote sensing inversion, the section location information and weight values ​​in this record will be read by the system and automatically loaded into the inversion algorithm as a "profile confidence mask" to control the fitting strength of the profile data in that section during the inversion process; in atmospheric analysis or numerical model training, this record is passed to the loss function module as a "sample confidence control factor" to reduce the backpropagation impact of that section of data on the model weights. For example, if a high-risk disturbance is identified in the 10.2km–10.6km altitude range in a radiosonde test, with a minimum weight of 0.21 and an average weight of 0.34, the system will generate a disturbance segment numbered D3. This segment will be bound to the spatial trajectory and total column count records of the 10.2–10.6km range, forming an error labeling package D3. In subsequent model training, the gradient influence weight of this data segment will be automatically reduced, and in the inversion process, it will be suppressed as a low-confidence interval and will not participate in the high-weight fitting process. Ultimately, all such error labeling units are output as a unified data structure, providing a standardized interface for subsequent algorithm calls and effectively improving the local robustness and overall error controllability of the entire radiosonde-inversion-modeling chain.

[0033] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention should still fall within the scope of the claims of the present invention.

Claims

1. A method for calculating the total column amount of ozone based on sounding observations, characterized in that, Includes the following steps: S1: Obtain the GPS positions at various altitudes during the launch of the weather balloon and construct a three-dimensional launch path; S2: Spatially overlay the three-dimensional ascent path with the air mass distribution layer in the meteorological reanalysis data of the corresponding time period to obtain the air mass type at each altitude point; S3: Based on the abrupt changes in air mass type, identify the height segments in ozone concentration changes that may be affected by lateral non-uniformity and generate profile confidence weights. S4: Introduce the profile confidence weight into the total column calculation process, perform weighted integration of ozone concentration at each height layer, and obtain the corrected total ozone column. S5: Output the lateral interference segment and its corresponding weight as an error-sensitive label to support subsequent model training or remote sensing inversion error correction.

2. The method for calculating the total column amount of ozone based on sounding observations according to claim 1, characterized by, The steps to obtain the GPS positions at various altitudes during the launch of the weather balloon and construct the three-dimensional launch path are as follows: A multi-band dual-channel GNSS positioning module is integrated into the sounding balloon payload system, which is used for main signal tracking and drift compensation signal acquisition, respectively, to obtain three-dimensional spatial coordinate data of the sounding balloon every second during the ascent process. The three-dimensional spatial coordinates include longitude, latitude and standard atmospheric altitude. The spatial coordinates collected every second are processed to smooth the path trajectory. An adaptive Kalman filter algorithm based on the acceleration rate of change constraint is used to identify and remove abnormal positioning drift points caused by sudden changes in wind speed or attitude turbulence, so as to obtain a three-dimensional trajectory sequence with good continuity. A three-dimensional launch path is constructed based on the trajectory sequence. The standard barosphere at each altitude point is bound to its corresponding GPS coordinates to form an elevation-position mapping set. The Earth ellipsoid reference model is then used to perform spatial correction on the mapping, resulting in a high-precision three-dimensional launch trajectory structure with a unified spatial benchmark.

3. The method for calculating the total column amount of ozone based on sounding observations according to claim 1, characterized by, The steps to spatially overlay the three-dimensional ascent path with the air mass distribution layer in the meteorological reanalysis data for the corresponding time period to obtain the air mass type at each altitude point are as follows: Based on the S-dimensional launch trajectory structure, the three-dimensional coordinate information of each trajectory point in the trajectory is used as a spatial query index to call the multi-layer air mass attribute layer in the meteorological reanalysis dataset that matches the sounding time period. To identify the air mass affiliation of 3D trajectory points, a layer fusion method based on voxel projection difference analysis is adopted: the 3D trajectory structure is projected onto the reanalysis layer coordinate system, and the trajectory points are dynamically identified as crossing the air mass boundary according to the spatial angle between the trajectory points and the boundaries of each grid cell, the direction of the boundary tangent vector, and the local gradient response characteristics. Each trajectory point is then classified into the characteristic air mass cell in which it is currently located, and grid domains that simultaneously satisfy the conditions of sudden wind direction change, sudden humidity change, and ozone concentration gradient reversal are preferentially selected as boundary identification factors. By binding the trajectory points that have been assigned to air mass types with their original pressure altitude and spatial coordinates, a structured air mass type label sequence is generated.

4. The method for calculating the total column amount of ozone based on sounding observations according to claim 1, characterized by, The steps for identifying high-altitude regions in ozone concentration changes that may be affected by lateral non-uniformity, based on abrupt changes in air mass type, and generating profile confidence weights include: Based on ozone concentration data at adjacent altitude points in the three-dimensional ascent trajectory, an ozone concentration change trend index is calculated. Calculate a boundary information inconsistency index based on the air mass type label sequence; Subtract the ozone concentration change trend index from the boundary information inconsistency index to obtain a potential horizontal error index; When the potential horizontal error index exceeds a preset threshold, identify the corresponding height section as a potential horizontal error section; Based on the potential horizontal error index corresponding to each height section, generate a profile reliability weight for each height point in the three-dimensional ascending trajectory.

5. The method for calculating the total column amount of ozone based on sounding observations according to claim 4, characterized by, The steps for calculating the ozone concentration change trend index based on the ozone concentration data of adjacent height points in the three-dimensional ascending trajectory are as follows: According to the order of height from low to high in the three-dimensional ascending trajectory, obtain the ozone concentration values corresponding to each height point in turn, and calculate the ozone concentration difference between adjacent two height points to represent the layer-by-layer change of ozone concentration in the vertical direction; for each ozone concentration difference, determine its change direction and convert it into a trend symbol, wherein a positive trend symbol corresponds to an ozone concentration difference that increases with height, a negative trend symbol corresponds to an ozone concentration difference that decreases with height, and a zero trend symbol corresponds to an ozone concentration difference that does not change, thereby forming a sequence of trend symbols arranged in height order; Select a continuous window containing several adjacent height points in the trend symbol sequence centered on the current height point, and count the length of the longest continuous segment of trend symbols with the same direction in the window to obtain the length of the longest continuous segment as a measure of the continuity of the change direction of the ozone concentration at the current height point; Within the same height range as the continuous window, count the maximum and minimum values of all ozone concentration differences in the window, and calculate the difference between them to represent the fluctuation range of the ozone concentration change amplitude in the height section; Jointly normalize the change direction continuity measure and the change amplitude fluctuation range, wherein the change direction continuity measure is scaled by the window length, and the change amplitude fluctuation range is normalized by the maximum ozone concentration change amplitude in the entire three-dimensional ascending trajectory, and the local trend stability value of the current height point is constructed through the product relationship of the two; Interval compression processing is performed on the local trend stability values corresponding to each height point to obtain the ozone concentration change trend index.

6. The method for calculating the total column amount of ozone based on sounding observations according to claim 4, characterized by, The steps for calculating the boundary information inconsistency index based on the air mass type label sequence are as follows: From the air mass type labels corresponding to each height point in the three-dimensional ascending trajectory, the air mass type label is a classification identification code generated after superimposing and analyzing multiple source meteorological layers; Iterate through each pair of adjacent height points to determine whether the air mass type labels of the two adjacent points change, if they change, mark the height point pair as a label mutation point, if they do not change, mark it as a label continuation point; For each label mutation point, the meteorological boundary attribute information corresponding to its spatial position is extracted, including: wind direction change angle, relative humidity jump amplitude and background ozone concentration gradient direction change. Respectively, if the wind direction change exceeds 45 degrees, the humidity jump exceeds 15%, and the ozone concentration gradient direction appears positive and negative reversal, then the boundary consistency state value of the label mutation point is 1; If only any two of the above three conditions are met, the state value is 0.66; If only one of them is met, the state value is 0.33; If none of them is met, the state value is 0; The sum of the boundary consistency state values of all label mutation points is counted, and then divided by the number of label mutation points to obtain the average boundary consistency value; The number of label mutation points is divided by the total number of adjacent height points to obtain the label mutation proportion value; The product of the above two values is taken as the boundary mutation significance; Further, the number of points in which the air mass label remains consistent but the absolute value of the ozone concentration difference is greater than the 95th percentile of the full trajectory concentration difference sequence is counted among all label continuation points, and the number is divided by the total number of label continuation points to obtain the structure drift mismatch degree; The boundary mutation significance and the structure drift mismatch degree are added and multiplied by one-half to obtain the boundary information inconsistency index.

7. The method of calculating the total column amount of ozone based on sounding observations according to claim 1, wherein, The steps for introducing the profile reliability weight into the column total calculation process, weighting the ozone concentration of each height layer, and obtaining the modified ozone column total are as follows: Based on the profile reliability weight sequence generated for each height point, the height points are divided into several continuous integration units according to their order in the ascending trajectory, each integration unit including a group of height points and their corresponding ozone concentration values and profile reliability weight values; For each integration unit, it is determined whether the profile reliability weight of all height points within it is greater than the preset full weight threshold value, if so, the complete height range of the integration unit is retained, and the corresponding ozone concentration value is multiplied by the corresponding physical thickness to participate in the column total calculation as a standard integration term; If there is a point in the integration unit whose profile reliability weight is lower than the preset local abnormal threshold value, the system will start the integration return mechanism: that is, a fixed height window is expanded upward and downward around the abnormal point, if the reliability in this interval shows a continuous downward trend, the data in this segment is excluded from the integration, if it shows intermediate abnormality or reliable at both ends, only the reliable segment at both ends is retained and the integration interval is compressed according to the reliability weight; For the remaining integration units, instead of direct weighting, a weight control function is introduced, which maps the profile reliability weight value to an integration coefficient adjustment factor. The adjustment method is as follows: when the weight value is greater than the high reliability threshold value, the integration coefficient remains unchanged; When the weight value is between the high reliability and the abnormal threshold value, the integration coefficient decays exponentially; when the weight value is lower than the abnormal threshold value, the integration coefficient is set to zero, and the height point is not included in the column total; The integration results of all integration units after elimination, compression and adjustment are accumulated as the modified ozone column total value.

8. The method of claim 1, wherein, The steps for outputting the lateral interference section and the corresponding weight as error sensitive annotations for supporting subsequent model training or remote sensing inversion error correction are as follows: Based on the identified potential lateral error sections and the corresponding profile confidence weight sequence, the height points in the three-dimensional ascending trajectory are reconstructed in sections. The intervals with profile confidence weights lower than the preset threshold in the continuous height points are merged into lateral interference sections, and a unique section identifier is generated for each lateral interference section. For each lateral interference section, the section start and end height, the minimum value, the average value, and the weight change gradient features of the profile confidence weight within the section are extracted, and the above information is packaged as a structured error annotation unit. The error annotation unit is bound with the corresponding three-dimensional ascending trajectory coordinates, sounding time stamp, and ozone column total amount calculation results to generate a traceable error sensitive annotation record, so that each annotated section can be located back to a specific height range in the original sounding path. The error sensitive annotation record is mapped to the model training interface and the remote sensing inversion correction interface respectively. In the model training process, the error sensitive annotation is used as sample confidence constraint information to reduce the influence weight of the corresponding height section data in the model parameter update. In the remote sensing inversion correction process, the error sensitive annotation is used as a profile confidence mask to limit the constraint strength of the inversion algorithm within the corresponding height section. The error sensitive annotation set containing the position of the lateral interference section, the corresponding profile confidence weight, and its annotation attribute is output as the unified input data for subsequent model training and remote sensing inversion error correction.

Images