An atmospheric boundary layer height acquisition method, device, medium and equipment

Abstract

This invention discloses a method, apparatus, medium, and device for obtaining atmospheric boundary layer height. The method includes: accessing multi-source observation data to generate observation records with quality indicators and initial observation uncertainties; performing spatiotemporal mapping to generate a registered profile sequence with an uncertainty field; generating a set of physical candidate heights in parallel based on at least one physical mechanism among thermal gradient, dynamic shear, and material distribution, and recording the variance of the source observations after interpolation for each physical candidate as the candidate uncertainty; constructing feature vectors for the pixels to be estimated; inputting the feature vectors into a first-layer candidate scorer to output a confidence score for each physical candidate, and inputting the feature vectors and confidence scores together into a second-layer heteroscedasticity regressor to output a point estimate and pixel-level uncertainty of the boundary layer height; using the point estimate and pixel-level uncertainty as observation terms, performing Kalman filtering assimilation with the background field, and outputting the atmospheric boundary layer height.

Description

Technical Field

[0001] This invention relates to the field of atmospheric altitude calculation technology, and more specifically, to a method, apparatus, medium, and device for obtaining atmospheric boundary layer altitude. Background Technology

[0002] The atmospheric boundary layer height (PBLH or Mixing Layer Height, MLH) characterizes the mixing top depth between the Earth's surface and the free atmosphere. It is a key physical quantity for numerous meteorological and environmental applications, including pollutant diffusion, heat and moisture fluxes, turbulent mixing, and numerical weather prediction. Traditional direct radiosondes (radiosondes, tethered balloons, aircraft) can provide high-precision temperature, humidity, and wind profiles, but their spatial coverage and frequency are insufficient, and maintenance costs are high. Ground-based remote sensing (such as LiDAR, wind profiler radar / RASS) has high temporal resolution but is affected by weather / topography, resulting in limited spatial representativeness. Satellite observations (such as GPS-RO) have wide coverage and rich vertical information, but their spatiotemporal refresh rate and near-surface inversion accuracy are limited. A single observation method cannot simultaneously meet the requirements of high precision, high coverage, and high spatiotemporal resolution. Therefore, an engineered method that integrates multiple sources and quantifies uncertainties is needed to provide reliable regional / real-time MLH products.

[0003] Limitations of existing technologies: Radiosonde observation: high precision but low frequency, poor coverage, and high cost, making it difficult to meet real-time and regional requirements; Ground-based remote sensing: continuous monitoring is possible but fails or produces significant errors in rainfall, low visibility, or complex terrain; Satellite observation: wide coverage but inversion errors are large under near-surface vertical resolution or strong convection conditions; Fusion methods: most are empirical weighting or black-box regression, lacking end-to-end uncertainty propagation and uncertainty-driven assimilation weighting mechanisms; Real-time performance and availability: existing schemes struggle to provide traceable backoff strategies and confidence outputs when observations are missing or abnormal. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method, apparatus, medium, and device for obtaining atmospheric boundary layer height.

[0005] According to one aspect of the present invention, a method for obtaining atmospheric boundary layer height is provided, comprising: Parallel access to multi-source observation data, signal-to-noise ratio detection and anomaly removal for each observation record, generating observation records with quality indicators and initial observation uncertainties; The observation records with initial observation uncertainty are mapped to a unified output time and a unified vertical grid. Uncertainty is propagated and updated synchronously during the interpolation process to generate a registered profile sequence with an uncertainty field. On the profile sequence, a set of physical candidate heights is generated in parallel based on at least one physical mechanism among thermal gradient, dynamic shear and material distribution, and the variance of its source observation after interpolation is recorded as the candidate uncertainty for each physical candidate. Construct a feature vector for the pixel to be estimated. The feature vector includes at least a set of physical candidate heights, candidate uncertainties, and observation metadata. The feature vector is input into the first-layer candidate scorer, which outputs the confidence score of each physical candidate. The feature vector and the confidence score are then input into the second-layer heteroscedasticity regressor, which outputs the point estimate of the boundary layer height and the pixel-level uncertainty. The point estimate and pixel-level uncertainty are used as observation terms and assimilated with the background field by Kalman filtering. The observation error covariance matrix is ​​constructed based on the pixel-level uncertainty and representative error. The output is the analysis field and analysis covariance, where the analysis field is the atmospheric boundary layer height. For each pixel in the analysis field, an innovation test and an uncertainty threshold test are performed. If the test fails, a backoff strategy is triggered. The backoff strategy prioritizes the physical candidate with the highest confidence score in the physical candidate height set as the replacement, and records the replacement source and triggering reason.

[0006] According to another aspect of the present invention, an atmospheric boundary layer height acquisition device is provided, comprising: The first generation module is used to access multi-source observation data in parallel, perform signal-to-noise ratio detection and anomaly removal on each observation record, and generate observation records with quality indicators and initial observation uncertainties. The second generation module is used to map observation records with initial observation uncertainty to a unified output time and a unified vertical grid, synchronously propagate and update uncertainty during the interpolation process, and generate a registered profile sequence with uncertainty field. The third generation module is used to generate a set of physical candidate heights in parallel on the profile sequence based on at least one physical mechanism among thermal gradient, dynamic shear and material distribution, and to record the variance of the source observation after interpolation for each physical candidate as the candidate uncertainty. The construction module is used to construct a feature vector for the pixel to be estimated. The feature vector includes at least a set of physical candidate heights, candidate uncertainties, and observation metadata. The output module is used to input the feature vector into the first-layer candidate scorer, output the confidence score of each physical candidate, and input the feature vector and confidence score together into the second-layer heteroscedasticity regressor, outputting the point estimate of the boundary layer height and the pixel-level uncertainty. The assimilation module is used to treat the point estimates and pixel-level uncertainties as observations and perform Kalman filtering assimilation with the background field. The observation error covariance matrix is ​​constructed based on the pixel-level uncertainty and representative error. The output is the analysis field and analysis covariance, where the analysis field is the atmospheric boundary layer height. The inspection module is used to perform innovation inspection and uncertainty threshold inspection on each pixel in the analysis field. If the inspection fails, the backoff strategy is triggered. The backoff strategy prioritizes the physical candidate with the highest confidence score in the physical candidate height set as the replacement, and records the replacement source and triggering reason.

[0007] According to another aspect of the present invention, a computer-readable storage medium is provided, the storage medium storing a computer program for performing the methods described in any of the above aspects of the present invention.

[0008] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the method described in any of the preceding aspects of the present invention.

[0009] Therefore, this invention proposes an end-to-end closed-loop system: "Physical Candidates (using a small number of interpretable physical candidates as learner input) → Candidate Scoring → Heteroscedastic Regression (output point estimation and pixel-level uncertainty) → Uncertainty-Driven Kalman Assimilation (constructing the observation covariance matrix R) → Dynamic Spatiotemporal Adaptive Weights and Backoff Strategy," thus achieving a balance between physical interpretability, data-driven accuracy, and mathematical rigor of assimilation. This system can provide high spatiotemporal resolution, confident MLH products, and maintain product continuity and traceability through candidate backoff and background regression in the event of observational anomalies or missing data, significantly outperforming traditional single-source or empirically weighted methods. Attached Figure Description

[0010] Exemplary embodiments of the present invention can be more fully understood by referring to the following figures: Figure 1 This is a flowchart illustrating an exemplary embodiment of the method for obtaining atmospheric boundary layer height provided by the present invention. Figure 2 This is a schematic diagram of the structure of an atmospheric boundary layer height acquisition device provided in an exemplary embodiment of the present invention; Figure 3 This is the structure of an electronic device provided in an exemplary embodiment of the present invention. Detailed Implementation

[0011] Hereinafter, exemplary embodiments according to the present invention will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments described herein.

[0012] It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of the invention.

[0013] Those skilled in the art will understand that the terms "first," "second," etc., in the embodiments of the present invention are only used to distinguish different steps, devices, or modules, and do not represent any specific technical meaning, nor do they indicate a necessary logical order between them.

[0014] It should also be understood that in the embodiments of the present invention, "multiple" can refer to two or more, and "at least one" can refer to one, two or more.

[0015] It should also be understood that any component, data or structure mentioned in the embodiments of the present invention can generally be understood as one or more unless explicitly defined or given contrary instructions in the context.

[0016] Furthermore, the term "and / or" in this invention is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this invention generally indicates that the preceding and following related objects have an "or" relationship.

[0017] It should also be understood that the description of the various embodiments in this invention emphasizes the differences between the various embodiments, and the similarities or similarities can be referred to each other. For the sake of brevity, they will not be described in detail.

[0018] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0019] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit the invention or its application or use.

[0020] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, they should be considered part of the specification.

[0021] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0022] The embodiments of this invention can be applied to electronic devices such as terminal devices, computer systems, and servers, and can operate together with a wide range of other general-purpose or special-purpose computing system environments or configurations. Well-known examples of terminal devices, computing systems, environments, and / or configurations suitable for use with electronic devices such as terminal devices, computer systems, and servers include, but are not limited to: personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, and distributed cloud computing environments including any of the above systems, etc.

[0023] Electronic devices such as terminal devices, computer systems, and servers can be described in the general context of computer system executable instructions (such as program modules) executed by a computer system. Typically, program modules can include routines, programs, object programs, components, logic, data structures, etc., which perform specific tasks or implement specific abstract data types. Computer systems / servers can be implemented in distributed cloud computing environments, where tasks are executed by remote processing devices linked through communication networks. In distributed cloud computing environments, program modules can reside on local or remote computing system storage media, including storage devices.

[0024] Exemplary methods Figure 1 This is a schematic flowchart of an exemplary embodiment of the present invention for obtaining atmospheric boundary layer height. This embodiment can be applied to electronic devices, such as… Figure 1 As shown, the atmospheric boundary layer height acquisition method 100 includes the following steps: Step 101: Parallel access of multi-source observation data, signal-to-noise ratio detection and anomaly removal for each observation record, and generation of observation records with quality indicators and initial observation uncertainty; Step 102: Map the observation records with initial observation uncertainty to a unified output time and a unified vertical grid, and simultaneously propagate and update the uncertainty during the interpolation process to generate a registered profile sequence with an uncertainty field. Step 103: On the profile sequence, based on at least one physical mechanism among thermal gradient, dynamic shear and material distribution, a set of physical candidate heights is generated in parallel, and the variance of the source observation after interpolation is recorded as the candidate uncertainty for each physical candidate. Step 104: Construct a feature vector for the pixel to be estimated. The feature vector includes at least a set of physical candidate heights, candidate uncertainties, and observation metadata. Step 105: Input the feature vector into the first-layer candidate scorer, output the confidence score of each physical candidate, and input the feature vector and confidence score together into the second-layer heteroscedasticity regressor, output the point estimate of the boundary layer height and the pixel-level uncertainty. Step 106: The point estimate and pixel-level uncertainty are taken as observation terms and assimilated with the background field by Kalman filtering. The observation error covariance matrix is ​​constructed based on the pixel-level uncertainty and representative error. The analysis field and analysis covariance are output, where the analysis field is the atmospheric boundary layer height. Step 107: Perform innovation test and uncertainty threshold test on each pixel in the analysis field. If the test fails, trigger the backoff strategy. The backoff strategy prioritizes the physical candidate with the highest confidence score in the physical candidate height set as the replacement, and records the replacement source and triggering reason.

[0025] Specifically, addressing the technical problems existing in the background technology, this invention innovatively proposes an end-to-end closed-loop system of "physical candidates (using a small number of interpretable physical candidates as learner input) → candidate scoring → heteroscedasticity regression (output point estimation and pixel-level uncertainty) → uncertainty-driven Kalman assimilation (constructing the observation covariance matrix R) → dynamic spatiotemporal adaptive weights and backoff strategy," thereby achieving a balance between physical interpretability, data-driven accuracy, and mathematical rigor of assimilation. This system can provide high spatiotemporal resolution, MLH products with confidence, and maintain product continuity and traceability through candidate backoff and background regression in the case of observational anomalies or missing data, significantly outperforming traditional methods of single-source or empirical weighting. The specific implementation steps are as follows: 1. Data Acquisition and Preliminary Quality Control The system receives in parallel backscattering profile β(z,t) from ground-based lidar, horizontal wind components u(z,t) and v(z,t) from boundary layer wind profiler radar / RASS, and available temperature / potential temperature T(z,t) / θ(z,t), as well as refractive index or near-surface temperature and humidity profiles from satellite GPS-RO as vertical background information. Each observation undergoes signal-to-noise ratio (SNR) detection, timestamp consistency verification, and anomaly removal upon entry into the database. A quality flag (QC_flag), observation time uncertainty (Δt_source), and initial observation uncertainty (σ_obs_source) are also attached to the record. Simultaneously, the variance of each observation is estimated and saved as metadata for subsequent interpolation and weighting. Low SNR, signal loss, or other obvious anomalies are flagged or excluded to ensure that data entering the fusion link carries quantifiable confidence information, supporting uncertainty-driven subsequent processing.

[0026] 2. Temporal-spatial registration and vertical remeshing Map heterogeneous observations to a unified output time. With uniform vertical layer Temporal interpolation employs exponentially decaying time weights to reduce the impact of long-term observations on real-time estimation. Spatially, point observations are weighted by distance, and extrapolation outside the observation coverage area is prohibited. Vertically, linear or cubic spline interpolation is used to insert data into a preset vertical grid (e.g., 0–3000 m, every 25–100 m), with uncertainties propagated and updated synchronously during registration / interpolation. The output is a registered profile sequence with an uncertainty field, which serves as input for candidate generation and feature construction.

[0027] 3. Physics candidate generation (a small number of interpretable candidates) A small number of representative sets of physical candidate altitudes are generated in parallel on the registered temperature and humidity wind and ground-based lidar echo profiles. The model is based on three main clues: thermal, dynamic, and material distribution. (1) Where the potential temperature or virtual potential temperature gradient is significant (thermal candidate); (2) Where the gradient Richardson number first exceeds the critical value (dynamic candidate); (3) Where the first derivative of the backscattering of the ground-based lidar is maximum (material candidate). The variance of the interpolated source observations for each candidate record is used as the uncertainty of that candidate. The "candidate-source uncertainty" is used to replace the massive number of unstructured candidates, thereby reducing the complexity of the model input while maintaining physical interpretability, and providing traceable alternatives for subsequent scoring and backoff.

[0028] Formula (Potential Temperature Calculation and Ground-Based LiDAR Gradient Candidate Localization): In the formula, Potential temperature (K); Temperature (K); : Air pressure (hPa or Pa); Reference air pressure (usually 1000 hPa); : Gas constant for dry air (≈ 287.05 J·kg⁻¹·K⁻¹); Specific heat of dry air at constant pressure (≈ 1004 J·kg⁻¹·K⁻¹); LiDAR backscattering coefficient (m⁻¹ sr⁻¹ or dimensionless); : β is the derivative of height (m⁻² sr⁻¹); The height (m) corresponding to the maximum gradient is used as a material candidate.

[0029] 4. Feature engineering (constructing multimodal input vectors) Construct feature vectors for the pixels to be estimated This includes: virtual temperature values ​​at several representative height levels. Alternatively, the model may use its gradient, wind shear index (characterizing dynamic shear), maximum gradient value and corresponding height of radar echoes, MLH and background uncertainty of GPS-RO background, physical candidate set and its source variance, and observation metadata (such as time difference Δt, horizontal distance d, QC_flag). Continuous quantities are standardized, and missing profiles are imputed with uncertainty using background values ​​and background variance. Uncertainty of candidate sources is used as an explicit feature so that the model can identify the confidence differences of each observation component during training and inference, improving robustness under conditions of sparse / non-uniform observations.

[0030] 5. Hybrid Recognizer: Candidate Scores and Heteroscedastic Regression A two-layer recognizer is used: the first layer is a candidate scorer (e.g., random forest, XGBoost), which scores each physical candidate... Based on features Output credibility score The second layer is a heteroscedastic regressor (e.g., a deep neural network or Bayesian regression), to... With candidate scores As input, output MLH point estimate With pixel-level uncertainty During training, heteroscedasticity negative log-likelihood loss is used to enable the model to simultaneously learn and predict the mean and variance, thus allowing the output variance to reflect the input combination conditions and model confidence. The candidate scorer provides an interpretability path, and the heteroscedasticity regressor provides observational uncertainty that can be directly used for assimilation, achieving synergy between physical rules and data-driven approaches.

[0031] Formula (Confidence Score and Heteroscedasticity Loss): Parameter explanation: : The credibility score of the i-th candidate, ranging from [0,1]; Candidate scorer models (such as random forests, gradient boosting trees, etc.); Heteroscedasticity negative log-likelihood loss is used to train the regressor. : The true value MLH(m) of training sample n; : The mean of the model predictions (m); : Variance of model predictions (m²).

[0032] 6. Kalman filtering or ensemble Kalman filtering for time-series assimilation Point estimation of the output of the hybrid recognizer With output variance Considered as an observation term and background field Assimilation to obtain the analytical field Analysis of covariance For linear or linearizable systems, the Kalman gain is calculated using the Kalman update. And update the status: If the system has high dimensionality or significant nonlinearity, then ensemble Kalman filtering is used to approximate the system through ensemble samples. The samples are then updated using the Monte Carlo method. The key point lies in the observation error covariance matrix. The construction must directly contain The observation representativeness error is used to ensure that the assimilation weights are based on quantifiable uncertainty rather than empirical weights.

[0033] Formula (Kalman gain and analysis update): Parameter explanation: Background (prior) state vector (meshized MLH, unit m); Background covariance matrix (m²); : Observation operator (the mapping matrix from state to observation space); Observation vector (hybrid recognizer) (unit: m) : Observation error covariance matrix, entries can be derived from Combined with representative error (m²); Kalman gain matrix; : Analyze the (assimilated) state vector (m).

[0034] 7. Dynamic Spatiotemporal Registration and Adaptive Weights To measure the effective contribution of each observation to the target grid point, a unified adaptive weighting function is defined to encode time delay, spatial distance, and uncertainty in a unified manner: The adaptive weight is calculated for observation i. Then normalize This weight can be used to combine multiple observations at the same time to form a single observation term, or to map uncertainty to observation covariance. (Larger weights correspond to smaller variance). The weighting function automatically balances temporal availability, geographical representativeness, and observation / model confidence; if an observation is unavailable, its weights are reset to zero and it is renormalized to ensure the robustness of the assimilation process to unavailable observations.

[0035] Parameter explanation: Time decay factor: Spatial attenuation factor: Uncertainty inverse factor: multiplied by . : Unnormalized weights (dimensionless) of observation i. : The time difference (in minutes) between observation i and the target time. : Time scale parameter (e.g., 30 minutes). : Horizontal distance (in km or m) from observation i to the target grid point. Spatial scale parameters (to control spatial attenuation, recommended 5–20 km). : The estimated variance (m²) of the observation or model output for observation i.

[0036] 8. Quality control, backoff strategies, and uncertainty labeling After obtaining the analysis field, innovation tests and uncertainty threshold checks are performed on each pixel: the difference between observation and background is calculated. Calculate the standard deviation between observations and background. Compare; if (usually taken) or model output uncertainty Exceeding the preset threshold If the cell is deemed unsatisfactory, it is marked as low confidence and triggers a backoff strategy—prioritizing the highest-scoring physical candidate from the candidate set as a replacement, and recording the source and uncertainty of the replacement. If the candidate is unavailable, the background field value is returned. All backoff records and triggering reasons are saved as product metadata with the final output for post-audit and scientific verification, ensuring that the product remains engineering-ready and traceable under abnormal conditions.

[0037] Parameter explanation: :Difference between observation and background. Observation vector (hybrid recognizer) (unit: m). Background (prior) state vector (meshable MLH, unit m). : Observation operator (the mapping matrix from state to observation space). : Differential threshold factor (usually 3, corresponding to the 3σ test). : Observation error covariance matrix, entries can be derived from Combined with representative error (m²). Background covariance matrix (m²). : Observational standard deviation and background standard deviation, derived from background covariance With observation covariance Jointly decided (unit: m).

[0038] 9. Model training, online self-adaptation, and product output In the offline phase, a training set is constructed using high-quality radiosonde and long-term ground-based lidar / radar records from multiple seasons and weather conditions. Candidate scorers and heteroscedastic regressors are trained separately. Hyperparameters are selected through cross-validation, and a loss function with an uncertainty term ensures that the model output can be directly used for assimilation (i.e., simultaneously learning the mean and variance). In the runtime phase, small-batch online fine-tuning (low learning rate) is supported to address sensor drift or seasonal variations, and model version management and rollback are implemented. The system outputs standardized products at a set frequency (e.g., hourly). Each pixel includes: analysis value MLH(m), pixel standard deviation or variance (m₁ or m²), quality label, list of observation sources used, and timestamp. Historical outputs and logs are also saved for posterior validation and model retraining.

[0039] Formula (illustrative of training objective function): In the formula, : The set of model parameters (including the weights of candidate scorers and regressors). MLH(m) is the true value of training sample n. : The model's predicted mean (m) for sample n; : The model's prediction variance for sample n (m²).

[0040] Therefore, this invention proposes an end-to-end closed-loop system: "Physical Candidates (using a small number of interpretable physical candidates as learner input) → Candidate Scoring → Heteroscedastic Regression (output point estimation and pixel-level uncertainty) → Uncertainty-Driven Kalman Assimilation (constructing the observation covariance matrix R) → Dynamic Spatiotemporal Adaptive Weights and Backoff Strategy," thus achieving a balance between physical interpretability, data-driven accuracy, and mathematical rigor of assimilation. This system can provide high spatiotemporal resolution, confident MLH products, and maintain product continuity and traceability through candidate backoff and background regression in the event of observational anomalies or missing data, significantly outperforming traditional single-source or empirically weighted methods.

[0041] Exemplary device Figure 2 This is a schematic diagram of the atmospheric boundary layer height acquisition device provided in an exemplary embodiment of the present invention. Figure 2 As shown, the device 200 includes: The first generation module 210 is used to access multi-source observation data in parallel, perform signal-to-noise ratio detection and anomaly removal on each observation record, and generate observation records with quality indicators and initial observation uncertainties. The second generation module 220 is used to map observation records with initial observation uncertainty to a unified output time and a unified vertical grid, synchronously propagate and update uncertainty during the interpolation process, and generate a registered profile sequence with uncertainty field. The third generation module 230 is used to generate a set of physical candidate heights in parallel on the profile sequence based on at least one physical mechanism among thermal gradient, dynamic shear and material distribution, and to record the variance of the source observation after interpolation for each physical candidate as the candidate uncertainty. Construction module 240 is used to construct a feature vector for the pixel to be estimated. The feature vector includes at least a set of physical candidate heights, candidate uncertainties, and observation metadata. The output module 250 is used to input the feature vector into the first-layer candidate scorer, output the confidence score of each physical candidate, and input the feature vector and confidence score together into the second-layer heteroscedasticity regressor, outputting the point estimate of the boundary layer height and the pixel-level uncertainty. The assimilation module 260 is used to treat the point estimate and pixel-level uncertainty as observation terms and perform Kalman filtering assimilation with the background field. The observation error covariance matrix is ​​constructed based on the pixel-level uncertainty and representative error. The output is the analysis field and analysis covariance, where the analysis field is the atmospheric boundary layer height. The inspection module 270 is used to perform innovation inspection and uncertainty threshold inspection on each pixel in the analysis field. If the inspection fails, an avoidance strategy is triggered. The avoidance strategy prioritizes the physical candidate with the highest confidence score in the physical candidate height set as the replacement, and records the replacement source and triggering reason.

[0042] Exemplary electronic devices Figure 3 This is the structure of an electronic device provided in an exemplary embodiment of the present invention. For example... Figure 3 As shown, the electronic device 30 includes one or more processors 31 and memory 32.

[0043] The processor 31 may be a central processing unit (CPU) or other form of processing unit with data processing and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions.

[0044] The memory 32 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 31 may execute the program instructions to implement the methods of the software programs of the various embodiments of the present invention described above, and / or other desired functions. In one example, the electronic device may also include an input device 33 and an output device 34, these components being interconnected via a bus system and / or other forms of connection mechanisms (not shown).

[0045] In addition, the input device 33 may also include, for example, a keyboard, a mouse, etc.

[0046] The output device 34 can output various information to the outside. The output device 34 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0047] Of course, for the sake of simplicity, Figure 3 Only some of the components of this electronic device relevant to the present invention are shown, omitting components such as buses, input / output interfaces, etc. In addition, the electronic device may include any other suitable components depending on the specific application.

[0048] Exemplary computer program products and computer-readable storage media In addition to the methods and apparatus described above, embodiments of the present invention may also be computer program products, which include computer program instructions that, when executed by a processor, cause the processor to perform the steps in the methods according to various embodiments of the present invention described in the "Exemplary Methods" section above.

[0049] The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of the present invention. The programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on the user's computing device, partially on the user's computing device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0050] Furthermore, embodiments of the present invention may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps of the methods according to various embodiments of the present invention described in the "Exemplary Methods" section above.

[0051] The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, system, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0052] The basic principles of the present invention have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in the present invention are merely examples and not limitations, and should not be considered as essential features of each embodiment of the present invention. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the present invention to the necessity of employing the aforementioned specific details.

[0053] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For system embodiments, since they largely correspond to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0054] The block diagrams of devices, systems, devices, and systems involved in this invention are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, systems, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0055] The methods and systems of the present invention may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the methods is for illustrative purposes only, and the steps of the methods of the present invention are not limited to the order specifically described above unless otherwise specifically stated. Furthermore, in some embodiments, the present invention may also be implemented as a program recorded on a recording medium, the program comprising machine-readable instructions for implementing the methods according to the present invention. Thus, the present invention also covers recording media storing programs for performing the methods according to the present invention.

[0056] It should also be noted that in the systems, apparatus, and methods of the present invention, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered equivalents of the present invention. The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the invention. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of the invention. Therefore, the invention is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0057] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the invention to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for obtaining atmospheric boundary layer height, characterized in that, include: Parallel access to multi-source observation data, signal-to-noise ratio detection and anomaly removal for each observation record, generating observation records with quality indicators and initial observation uncertainties; The observation records with the initial observation uncertainty are mapped to a unified output time and a unified vertical grid. During the interpolation process, the uncertainty is propagated and updated synchronously to generate a registered profile sequence with an uncertainty field. On the profile sequence, a set of physical candidate heights is generated in parallel based on at least one physical mechanism among thermal gradient, dynamic shear and material distribution, and the variance of the source observation after interpolation is recorded as the candidate uncertainty for each physical candidate. A feature vector is constructed for the pixel to be estimated, and the feature vector includes at least the set of physical candidate heights, the candidate uncertainty, and observation data. The feature vector is input into the first-layer candidate scorer, which outputs the confidence score of each physical candidate. The feature vector and the confidence score are then input into the second-layer heteroscedasticity regressor, which outputs the point estimate of the boundary layer height and the pixel-level uncertainty. The point estimate and the pixel-level uncertainty are used as observation terms and assimilated with the background field by Kalman filtering. The observation error covariance matrix is ​​constructed based on the pixel-level uncertainty and the representative error. The analysis field and analysis covariance are output, where the analysis field is the atmospheric boundary layer height. Each pixel in the analysis field is subjected to innovation test and uncertainty threshold test. If the test fails, a backoff strategy is triggered. The backoff strategy prioritizes the physical candidate with the highest confidence score in the physical candidate height set as the replacement, and records the replacement source and triggering reason.

2. The method according to claim 1, characterized in that, The multi-source observation data includes at least one of the following: backscattering profiles from ground-based lidar, horizontal wind components provided by boundary layer wind profiler radar, and refractive index or temperature and humidity profiles from satellite GPS-RO; the initial observation uncertainty is estimated and saved as metadata when the data is entered into the database. The set of physical candidate heights includes: Thermal candidates: generated based on locations with significant potential temperature or virtual potential temperature gradients; Dynamic candidates: generated when the gradient Richardson number first exceeds the critical value; Material candidates: generated based on the location of the maximum first derivative of backscattering from ground-based lidar; The formula for calculating the potential temperature in the thermal candidate is as follows: In the formula, Potential temperature; :temperature; Air pressure; Reference air pressure; : Gas constant for dry air; Specific heat at constant pressure of dry air; The calculation expression for the ground-based lidar gradient candidate localization of the material candidates is as follows: In the formula, LiDAR backscattering coefficient; : The derivative of β with respect to altitude; : The height corresponding to the maximum gradient, used as a material candidate, z: height.

3. The method according to claim 1, characterized in that, The observation records with the initial observation uncertainty are mapped to a unified output time and a unified vertical grid, including: spatiotemporal registration and regrinding, time interpolation using exponential decay time weights, spatial interpolation using distance weights and prohibiting extrapolation outside the observation coverage area, vertical interpolation using linear or cubic spline interpolation to a preset vertical grid, and simultaneous propagation and updating of uncertainty during the interpolation process.

4. The method according to claim 1, characterized in that, The feature vector also includes the virtual potential temperature or its gradient of the representative height layer, wind shear index, maximum gradient value and corresponding height of radar echo, boundary layer height and background uncertainty of GPS-RO background, and time difference, horizontal distance and quality indicator in the observation metadata.

5. The method according to claim 1, characterized in that, The heteroscedastic regressor is trained using a heteroscedastic negative log-likelihood loss function, enabling the model to simultaneously learn to predict the mean and variance. The output pixel-level uncertainty reflects the confidence of the input combination conditions and the model. The credibility score P i and heteroscedasticity loss The calculation expression is: In the formula, : The credibility score of the i-th candidate, ranging from [0,1]; Physical candidates; : Feature vector; Candidate scorer model; Heteroscedasticity negative log-likelihood loss is used to train the regressor. The true value of training sample n; The mean of the model predictions; : Variance of model predictions; Point estimate; Pixel-level uncertainty.

6. The method according to claim 1, characterized in that, The point estimate and the pixel-level uncertainty are used as observation terms and assimilated with the background field using Kalman filtering, including: Kalman filtering or ensemble Kalman filtering is used to assimilate the point estimates and pixel-level uncertainties as observation terms with the background field using Kalman filtering. The construction of the observation error covariance matrix R directly includes the pixel-level uncertainties and observation representativeness errors to ensure that the assimilation weights are based on quantifiable uncertainties rather than empirical weights. For linear or linearizable systems, the Kalman gain is calculated using the Kalman update. And update the status: If the system has high dimensionality or significant nonlinearity, then ensemble Kalman filtering is used to approximate the system through ensemble samples. And update the samples using the Monte Carlo method, where Kalman gain The expression for state update is: In the formula, Background state vector; Background covariance matrix; : Observation operator; : Observation vector; : Observation error covariance matrix, entries can be derived from Combined with representative error; Kalman gain matrix; : Analyze the (assimilated) state vector (m).

7. The method according to claim 1, characterized in that, The method further includes: Before the time series assimilation step, an adaptive weight is calculated for each observation and normalized to obtain a normalized adaptive weight. The adaptive weight is encoded by integrating time delay, spatial distance and uncertainty, and is used to merge multiple observations or map observation covariance. The adaptive weights and the normalized adaptive weights The calculation formula is: In the formula, the time decay factor is: Spatial attenuation factor: Uncertainty inverse factor: multiplied by ; Observation i Unnormalized adaptive weights; : The time difference between observation i and the target time; Time scale parameters; Observation i Horizontal distance to the target grid point; Spatial scale parameters; : Observation or model output for observation i The estimated variance.

8. The method according to claim 1, characterized in that, The innovation test is based on the difference between observation and background. Standard deviation of observation and background If a comparison is made, Or model output uncertainty Exceeding the preset threshold If the test fails, the backoff strategy is triggered. :Difference between observation and background; : Observation vector; Background state vector; : Observation operator; : Differential threshold factor; : Observation error covariance matrix, entries can be derived from Combined with representative error; Background covariance matrix; : Observational standard deviation and background standard deviation, derived from background covariance With observation covariance A joint decision.

9. The method according to claim 1, characterized in that, The method further includes: In the offline phase, a training set is constructed using high-quality radiosonde data and long-term ground-based observations to train the candidate scorer and the heteroscedasticity regressor, respectively. In the operational phase, small-batch online fine-tuning is supported to adapt to sensor drift or seasonal changes.

10. A device for obtaining atmospheric boundary layer height, characterized in that, include: The first generation module is used to access multi-source observation data in parallel, perform signal-to-noise ratio detection and anomaly removal on each observation record, and generate observation records with quality indicators and initial observation uncertainties. The second generation module is used to map the observation records with the initial observation uncertainty to a unified output time and a unified vertical grid, synchronously propagate and update the uncertainty during the interpolation process, and generate a registered profile sequence with an uncertainty field. The third generation module is used to generate a set of physical candidate heights in parallel on the profile sequence based on at least one physical mechanism among thermal gradient, dynamic shear and material distribution, and to record the variance of the source observation after interpolation for each physical candidate as the candidate uncertainty. A construction module is used to construct a feature vector for the pixel to be estimated, wherein the feature vector includes at least the set of physical candidate heights, the candidate uncertainty, and observation data. The output module is used to input the feature vector into the first-layer candidate scorer, output the confidence score of each physical candidate, and input the feature vector and the confidence score together into the second-layer heteroscedasticity regressor, outputting the point estimate of the boundary layer height and the pixel-level uncertainty. The assimilation module is used to take the point estimate and the pixel-level uncertainty as observations and perform Kalman filtering assimilation with the background field. The observation error covariance matrix is ​​constructed based on the pixel-level uncertainty and the representative error. The module outputs the analysis field and the analysis covariance, where the analysis field is the atmospheric boundary layer height. The inspection module is used to perform innovation inspection and uncertainty threshold inspection on each pixel in the analysis field. If the inspection fails, an avoidance strategy is triggered. The avoidance strategy prioritizes the physical candidate with the highest confidence score in the physical candidate height set as the replacement, and records the replacement source and triggering reason.

Images