Laser ozone lidar full profile adaptive correction method considering vertical spatial correlation

Abstract

The application discloses a laser ozone radar full profile adaptive correction method considering vertical space correlation, comprising the following steps: constructing a time-space matched laser radar and a high-precision reference observation sample set and performing quality control; for each target height layer, extracting observation data of adjacent layers in the vertical direction, and constructing a multi-dimensional feature vector containing vertical space correlation characteristics; establishing a multivariate linear regression model fusing a height drift correction term and a neighborhood coupling smoothing term, and solving the model parameters with clear physical meaning by using the least square method; and using the trained model to adaptively correct the original laser ozone radar data to be corrected. The application effectively overcomes the defects of traditional layered correction, such as high noise in high altitude, black box model unexplainable and the like, caused by ignoring vertical correlation, significantly improves the detection accuracy and data reliability of the laser ozone radar in the high altitude low signal-to-noise ratio area, and is suitable for atmospheric environment business monitoring.

Description

Technical Field

[0001] This invention relates to the field of atmospheric environment monitoring and meteorological detection technology, and more specifically, to an adaptive correction method for the full profile of laser ozone radar that takes into account vertical spatial correlation, which is suitable for improving the inversion accuracy and operational application quality of atmospheric ozone vertical distribution data. Background Technology

[0002] Ozone (O3) is a key trace gas in the stratosphere and troposphere, and its vertical distribution characteristics are directly related to ultraviolet radiation transport, atmospheric chemical processes, and climate change research. Currently, the main methods for obtaining ozone vertical profiles include ozone sounding and lidar detection. Ozone sounding offers high accuracy but low spatiotemporal resolution (typically twice daily); differential absorption lidar (DIAL) possesses high spatiotemporal resolution and continuous observation capabilities, making it an important tool for ground-based remote sensing monitoring.

[0003] However, laser ozone radar faces two major technological bottlenecks in practical applications: (1) Signal-to-noise ratio decreases with altitude: As the detection altitude increases, the laser echo signal weakens sharply, which amplifies the noise in the inversion algorithm, and the error of high-altitude data (usually >3km) increases significantly, and even non-physical negative values ​​or violent oscillations occur.

[0004] (2) Limitations of traditional correction methods: Hierarchical independent correction method: Current mainstream methods typically establish independent statistical correction models (such as univariate linear regression) for each altitude layer. This method disrupts the physical continuity of atmospheric ozone in the vertical direction, fails to utilize effective information from the neighborhood to suppress random noise in the current layer, resulting in extremely poor correction performance in high-altitude low signal-to-noise ratio regions, and uneven distribution of errors across the entire profile.

[0005] Black-box machine learning methods: While methods such as random forests and neural networks achieve high fitting accuracy on training sets, they lack physical interpretability and cannot meet the stringent requirements of meteorological operations for error tracing, quality control, and uncertainty assessment. Furthermore, black-box models have weak generalization capabilities, requiring massive amounts of new samples for retraining when migrating across seasons or sites, resulting in high maintenance costs.

[0006] Simple smoothing filtering: Traditional filtering methods such as moving average can suppress noise, but they can blur the true vertical gradient features (such as the ozone top and tropopause) and cannot correct systematic biases (such as background shift and high dependence error).

[0007] For a long time, those skilled in the art have generally believed that "layered independent modeling provides higher correction accuracy" because independent modeling can avoid "interference" between different altitude layers. This technical bias has led existing technologies to continue using the approach of layered independent modeling, while neglecting the continuous characteristics of the vertical distribution of atmospheric ozone. Therefore, there is an urgent need to develop an adaptive correction method for the full profile of laser ozone radar that can both suppress noise using vertical spatial correlation and correct system biases, and has clear physical meaning and is suitable for operational use. Summary of the Invention

[0008] The purpose of this invention is to propose an adaptive correction method for the full profile of a laser ozone radar that takes into account vertical spatial correlation. This method aims to solve the problems in the prior art, such as poor high-altitude correction effect due to ignoring vertical correlation, uninterpretable black box models, and the inability of traditional methods to simultaneously address noise reduction and bias correction.

[0009] To achieve the above objectives, in a first aspect, the present invention proposes an adaptive correction method for the full profile of a laser ozone radar that takes into account vertical spatial correlation, comprising: S1. Acquire spatiotemporally matched lidar ozone radar observation profile data and high-precision reference ozone profile data within historical time periods, and remove invalid data, outliers and excessively large data deviations to form sample pairs corresponding to each altitude layer. Each sample pair contains: the original lidar ozone concentration corresponding to an altitude layer, the reference ozone concentration corresponding to that altitude layer, and the altitude information corresponding to that altitude layer. S2. For each height layer, extract the original laser ozone concentration of that height layer, and extract the original laser ozone concentrations of the k adjacent lower and k adjacent upper height layers in the vertical direction. Combine the extracted original laser ozone concentrations with the height information of that height layer to form a vertical spatial correlation feature vector. Specifically, for the bottom boundary layer of the profile, only extract its k upper neighbor features, and for the top boundary layer of the profile, only extract its k lower neighbor features. S3. Establish a multiple regression model with the reference ozone concentration corresponding to the altitude layer as the dependent variable and the vertical spatial correlation feature vector corresponding to the same altitude layer as the independent variable. S4. Use the vertical spatial correlation feature vectors corresponding to each height layer and the reference ozone concentration corresponding to each height layer to form a training dataset, and train the multivariate regression model. Use the least squares method or regularized regression algorithm to solve the parameters of the multivariate regression model and obtain the optimal parameter combination. S5. For the original laser ozone radar data to be corrected, construct the vertical spatial correlation feature vector for each height layer according to the method in step S2, input it into the trained multivariate regression model, and calculate the corrected high-precision ozone vertical distribution data.

[0010] Optionally, the spatiotemporal matching in step S1 includes time synchronization and height matching; The time synchronization is such that the time difference between the reference observation transit period and the lidar observation period does not exceed a preset time threshold. The height matching is defined as the absolute value of the difference between the height of the reference observation data and the observation height of the lidar not exceeding a preset height threshold.

[0011] Optionally, in step S2, the formula for constructing the vertical spatial correlation feature vector is expressed as: X(h) = [h, , , ..., , , ..., ] Where X(h) is the vertical spatial correlation feature vector of the altitude layer, and h is the altitude of that altitude layer. This represents the original laser ozone concentration at that altitude. The original laser ozone concentration in the upper and lower neighboring height layers. i =1,2,...,k, where k is the number of one-sided neighborhood layers.

[0012] Optionally, in step S3, the mathematical expression of the multiple regression model is:

[0013] in, The corrected ozone concentration is represented by α; the intercept term is represented by β; the height drift coefficient is represented by γ0; and the response coefficient of this layer is represented by γ0. and These are the coupling weight coefficients for the i-th neighborhood of the lower and upper layers, respectively; i =1,2,...,k, where k is the number of one-sided neighborhood layers; The physical constraints of the above model parameters are as follows: The intercept term α is used to correct for system background bias. The height drift coefficient β is negative and is used to compensate for the underestimation of concentration caused by the attenuation of laser energy with increasing height, so as to correct the signal attenuation error that varies with height. The neighborhood coupling weight coefficient and The sum is not zero, and it is used to utilize the continuity of the vertical distribution of atmospheric ozone to perform low-pass filtering on high-frequency random noise at the current altitude layer in order to suppress random noise.

[0014] Optionally, in step S4, when the number of training samples is limited or multicollinearity exists, ridge regression or Lasso regression algorithms are used to optimize parameters and a regularization term is introduced to prevent model overfitting.

[0015] Optionally, a dynamic model update mechanism is also included: setting a rolling time window, periodically incorporating new spatiotemporally matched observation data, and repeatedly executing steps S1 to S4 to recalculate the model parameters to adapt to the performance drift, seasonal changes, and evolution of systematic errors caused by environmental factors of the lidar system.

[0016] Optionally, in step S1, the high-precision reference ozone profile data is derived from one or more fused data sources, including electrochemical ozone radiosonde, airborne ozone sensor, and satellite edge detection data.

[0017] Optionally, the value of the number of neighboring layers k in step S2 is dynamically adjusted based on the intensity of turbulent mixing and the vertical gradient in the atmospheric vertical structure: Within the atmospheric boundary layer, due to intense turbulent mixing, k takes values ​​of 2 to 4; In the free atmosphere, due to the large vertical gradient, k takes the value of 1 to 2.

[0018] Secondly, the present invention provides an electronic device, the electronic device comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect.

[0019] Thirdly, the present invention provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to perform the method described in the first aspect.

[0020] The beneficial effects of this invention are as follows: 1. Significantly improves the quality of high-altitude data: By introducing vertical neighborhood features and making full use of atmospheric continuity, random noise in low signal-to-noise ratio regions at high altitudes is effectively suppressed. Experiments show that at altitudes of 4500-5000m, the root mean square error (RMSE) improvement rate can reach over 68.8%, solving the problem of "high-altitude failure" in traditional layered correction.

[0021] 2. Combines denoising and bias correction functions: A single model simultaneously achieves system bias (intercept, height term) correction and random noise (neighborhood term) suppression, avoiding the error propagation problem caused by "smoothing first and then correcting" or "correcting first and then smoothing".

[0022] 3. Strong physical interpretability and adaptability to business operations: The model is based on multiple linear regression, and each coefficient has a clear physical meaning (such as the height attenuation coefficient corresponding to laser energy attenuation and the neighborhood coefficient corresponding to atmospheric continuity), which meets the meteorological business's specifications for data quality control and error tracing, and is superior to black-box machine learning models.

[0023] 4. Strong adaptability and generalization ability: Through a dynamic update mechanism, the model can automatically adapt to factors such as seasonal changes and instrument aging, and can be quickly deployed to new radar models or new sites without the need for massive historical data.

[0024] 5. Overcoming technical biases: This invention abandons the traditional perception that "layered independent modeling has higher accuracy" and proves that joint modeling of the entire profile can significantly improve the overall and local (especially high-altitude) correction accuracy while maintaining the vertical structural features, thus solving a long-standing technical problem in this field.

[0025] The present invention has other features and advantages, which will be apparent from or will be set forth in detail in the accompanying drawings and the following detailed description, which together serve to explain the particular principles of the invention. Attached Figure Description

[0026] The above and other objects, features and advantages of the present invention will become more apparent from the accompanying drawings, in which like reference numerals generally denote like parts.

[0027] Figure 1 This is a flowchart of an adaptive correction method for the full profile of a laser ozone radar that takes into account vertical spatial correlation, according to the present invention.

[0028] Figure 2 This is an analysis diagram showing the variation of model parameters with altitude in an embodiment of the present invention.

[0029] Figure 3 This is an analysis chart of the evaluation indicators for the model correction effect in an embodiment of the present invention.

[0030] Figure 4 This is a scatter plot comparing the effects before and after correction in an embodiment of the present invention. Detailed Implementation

[0031] To address the problems of poor high-altitude correction performance due to neglecting vertical correlation in existing technologies, uninterpretable black-box models, and the inability of traditional methods to simultaneously address noise reduction and bias correction, this invention discloses an adaptive correction method for the full profile of a laser ozone radar that takes into account vertical spatial correlation. This method first constructs a spatiotemporally matched laser radar and high-precision reference observation sample set and performs rigorous quality control. Then, for each target altitude layer, it extracts observation data from its vertically adjacent neighboring layers, constructing a multidimensional feature vector containing vertical spatial correlation features. Based on this, it establishes a multiple linear regression model that integrates altitude drift correction terms and neighborhood coupling smoothing terms, and uses the least squares method to solve for model parameters with clear physical meaning. Finally, it achieves adaptive correction of the full profile data through a dynamic update mechanism. This invention effectively overcomes the shortcomings of traditional layered correction methods, such as high high-altitude noise and uninterpretable black-box models due to neglecting vertical correlation, significantly improving the detection accuracy and data reliability of laser ozone radar in high-altitude low signal-to-noise ratio regions, and is suitable for operational atmospheric environmental monitoring.

[0032] The invention will now be described in more detail with reference to the accompanying drawings. While preferred embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0033] like Figure 1 As shown, an adaptive correction method for the full profile of a laser ozone radar considering vertical spatial correlation according to the present invention includes the following steps: S1. Acquire spatiotemporally matched lidar ozone radar observation profile data and high-precision reference ozone profile data within historical time periods, and remove invalid data, outliers and excessively large data deviations to form sample pairs corresponding to each altitude layer. Each sample pair contains: the original lidar ozone concentration corresponding to an altitude layer, the reference ozone concentration corresponding to that altitude layer, and the altitude information corresponding to that altitude layer. Specifically, this step constructs a high-quality training sample set: First, it collects spatiotemporally matched lidar observation profile data and high-precision reference ozone profile data from historical time periods. The high-precision reference ozone profile data comes from one or more fused data sources, including electrochemical ozone radiosonde, airborne ozone sensor, and satellite edge detection data. Spatiotemporal matching includes time synchronization and altitude matching. Time synchronization means that the time difference between the reference observation period and the lidar observation period does not exceed a preset time threshold. Altitude matching means that the absolute value of the difference between the altitude of the reference observation data and the lidar observation altitude does not exceed a preset altitude threshold. Then, through the implementation of strict multi-level data quality control (invalid value removal, IQR outlier detection, and relative deviation screening), a "raw observation-true value" paired sample library is constructed. Each sample pair contains: the raw lidar ozone concentration (raw observation) corresponding to a height layer, the reference ozone concentration (true value) corresponding to that height layer, and the altitude information corresponding to that height layer.

[0034] In this invention, "altitude layer" refers to the smallest sampling unit in the raw data of the lidar, divided according to a fixed vertical resolution. Each altitude layer corresponds to a range bin. For a vertically pointing lidar for ozone detection, its altitude... Take the center value of the sampling unit, vertical resolution This is the distance gate length. Specifically, if the vertical resolution of the lidar is... (In this embodiment) =7.5m), then the thickness of each height layer is The center height interval between adjacent height layers is In subsequent model training and correction, all operations are based on this original altitude layer scale to fully preserve the vertical structural details of atmospheric ozone and utilize its spatial continuity to suppress random noise.

[0035] S2. For each height layer, extract the original laser ozone concentration of that height layer, and extract the original laser ozone concentrations of the k adjacent lower and k adjacent upper height layers in the vertical direction. Combine the extracted original laser ozone concentrations with the height information of that height layer to form a vertical spatial correlation feature vector. Specifically, for the bottom boundary layer of the profile, only extract its k upper neighbor features, and for the top boundary layer of the profile, only extract its k lower neighbor features. Specifically, this step constructs a vertically correlated feature space: to overcome the limitations of single-layer modeling, for the target height layer h, not only is the original concentration L(h) of that layer extracted, but also the concentration data of its k adjacent layers above and below (k≥1). Utilizing the principle of continuity in the vertical distribution of atmospheric ozone, neighborhood information is used as a vertical spatial correlation feature to construct a multi-dimensional feature vector. The formula for constructing the vertical spatial correlation feature vector is expressed as: X(h) = [h, , , ..., , , ..., ] Where X(h) is the vertical spatial correlation feature vector of the altitude layer, and h is the altitude of that altitude layer. This represents the original laser ozone concentration at that altitude. The original laser ozone concentration in the upper and lower neighboring height layers. i =1,2,...,k, where k is the number of one-sided neighborhood layers.

[0036] In practice, the value of the number of neighboring layers k on one side can be dynamically adjusted according to the intensity of turbulent mixing and the vertical gradient in the atmospheric vertical structure. For example, in the atmospheric boundary layer, due to the strong turbulent mixing, the value of k is preferably 2 to 4; in the free atmosphere, due to the large vertical gradient, the value of k is preferably 1 to 2.

[0037] S3. Establish a multiple regression model with the reference ozone concentration corresponding to the altitude layer as the dependent variable and the vertical spatial correlation feature vector corresponding to the same altitude layer as the independent variable. Specifically, this step establishes a vertically coupled correction model: a multiple linear regression model is constructed with the reference ozone concentration corresponding to the altitude layer as the dependent variable and the vertical spatial correlation feature vector corresponding to the same altitude layer as the independent variable. The mathematical expression of this multiple regression model is as follows:

[0038] in, The corrected ozone concentration is represented by α; the intercept term is represented by β; the height drift coefficient is represented by γ0; and the response coefficient of this layer is represented by γ0. and These are the coupling weight coefficients for the i-th neighborhood of the lower and upper layers, respectively; i =1,2,...,k, where k is the number of one-sided neighborhood layers; The physical constraints of the above model parameters are as follows: The intercept term α is used to correct for system background bias; The height drift coefficient β is negative and is used to compensate for the underestimation of concentration caused by the attenuation of laser energy with increasing height, so as to correct the signal attenuation error that varies with height. Neighborhood Coupling Weight Coefficient and The sum is not zero, and it is used to utilize the continuity of the vertical distribution of atmospheric ozone to perform low-pass filtering on high-frequency random noise at the current altitude layer in order to suppress random noise.

[0039] The physical mechanism underlying the design of this model structure is as follows: Intercept term α: Corrects for instrument zero drift or background deviation; The height term β·h corrects for the systematic underestimation caused by the exponential decay of laser energy with height. This layer term γ0· L h Preserve the main signal response; Neighborhood Coupling Term : Utilize vertical correlation to perform adaptive smoothing filtering on random noise in the current layer through weighted neighborhood values.

[0040] S4. Use the vertical spatial correlation feature vectors corresponding to each height layer and the reference ozone concentration corresponding to each height layer to form a training dataset, and train the multivariate regression model. Use the least squares method or regularized regression algorithm to solve the parameters of the multivariate regression model and obtain the optimal parameter combination. Specifically, this step involves adaptive optimization of model parameters: a training dataset is constructed using the vertical spatial correlation feature vectors (independent variables) corresponding to each height layer and the reference ozone concentration (dependent variable) corresponding to each height layer, and the multivariate regression model constructed in step S3 is trained; the least squares method or regularization algorithm is used to solve for the optimal parameters of the model.

[0041] When the training sample size is limited or multicollinearity exists, ridge regression or Lasso regression algorithms are used to optimize parameters and regularization terms are introduced to prevent model overfitting.

[0042] S5. For the original laser ozone radar data to be corrected, construct the vertical spatial correlation feature vector for each height layer according to the method in step S2, input it into the trained multivariate regression model, and calculate the corrected high-precision ozone vertical distribution data.

[0043] Specifically, after constructing the vertical spatial correlation feature vectors of each height layer using the raw observation data of the real-time laser ozone radar according to the method in step S2, the trained multivariate regression model can be used to output a high-precision ozone profile after correction of all height layers in one go.

[0044] Preferably, this method also introduces a dynamic model update mechanism: by setting a rolling time window, new spatiotemporal matching observation data is periodically included, and the model parameters are updated using the latest observation data (steps S1 to S4 are repeated to recalculate the model parameters) to adapt to the performance drift, seasonal changes and system error evolution caused by environmental factors of the lidar system.

[0045] The present invention will now be described in more detail with reference to specific embodiments.

[0046] Example 1

[0047] This embodiment provides an adaptive correction method for the full profile of a laser ozone radar that takes into account vertical spatial correlation. (Refer to...) Figure 1 Specifically, it includes the following steps: Step 1: Data Acquisition and Preprocessing This embodiment selects observation data from a ground-based differential absorption lidar for the entire year of 2023. The lidar has a vertical resolution of 7.5 m, a temporal resolution of 1 hour, and a detection altitude range of 0.15 km to 5 km. The reference true values ​​are electrochemical ozone sounding data released at the same location during the same period, with ozone concentration units uniformly expressed as μg / m³.

[0048] Spatiotemporal matching: The time threshold is set to ±30 minutes, and the altitude threshold is set to ±3 m. That is, data in which the difference between the radiosonde's transit time and the lidar's observation time does not exceed 30 minutes, and the absolute value of the difference between the altitude reported by the radiosonde and the lidar's distance from the gate center does not exceed 3 m, will be retained.

[0049] Quality control: (1) Remove records where the laser inversion concentration is negative or zero; (2) Use the interquartile range (IQR) to remove outliers outside the range of Q1-1.5×IQR to Q3+1.5×IQR; (3) Remove gross error samples with a relative deviation >100%.

[0050] After the above processing, 12,735 valid sample pairs were finally obtained. Each sample pair includes: the altitude h of the altitude layer (taken as the center height of the gate at that distance), and the original laser ozone concentration of that altitude layer. L h and the reference ozone concentration (sounding value) corresponding to that altitude level. The dataset is divided into training and validation sets in a 7:3 ratio.

[0051] Step 2: Construct vertically correlated feature vectors

[0052] In this embodiment, "height layer" refers to the smallest sampling unit (i.e., a range gate) in the raw lidar data, divided according to a fixed vertical resolution (7.5 m). To facilitate the illustration of the overall trend, some figures are attached (…). Figures 2-4 Multiple adjacent original height layers are merged into a height interval at 500 m intervals for plotting, but this is only for visualization purposes. Model training and correction are performed on the original 7.5 m height layer.

[0053] Since the data in this embodiment covers the area from near the ground to 5 km, including the atmospheric boundary layer and the free atmosphere, the preferred value of k is 2 to 4 within the atmospheric boundary layer due to strong turbulent mixing; and the preferred value of k is 1 to 2 within the free atmosphere due to the large vertical gradient. For ease of explanation, this embodiment uses a single-sided neighboring layer number k=2 as an example.

[0054] For each height level in the training set Construct vertical spatial correlation feature vectors: For a layer located at an intermediate altitude in the free atmosphere, the eigenvector is X(h) = [h, , , ..., , , ..., ].

[0055] For the bottom layer, only the upper-layer neighborhood is considered, and the vertical spatial correlation feature vector is X(h) = [h, , , ]; For the top layer, only the lower-level neighborhood is considered, and the vertical spatial correlation feature vector is X(h) = [h, , , To ensure consistency in feature dimensions.

[0056] Thus, each height layer h corresponds to a vertical spatial correlation feature vector X(h), while the dependent variable remains the reference ozone concentration at that height layer. .

[0057] Step 3: Construction and Training of Vertically Coupled Correction Model

[0058] Establish a multiple linear regression model:

[0059] Using the vertical spatial correlation feature vector X(h) corresponding to all height layers in the training set as independent variables, The least squares method was used to solve for the model parameters as the dependent variable. Due to sufficient sample size and the absence of significant multicollinearity, no regularization term was used. Typical parameter results obtained from the model solution are as follows: α = 49.10 (background bias), β = -0.0029 (height attenuation correction), γ0 = -0.11 (current layer response), and neighborhood weight coefficients. and It exhibits an alternating positive and negative distribution, reflecting its filtering characteristics and effectively suppressing random noise.

[0060] Figure 2The model parameters are shown to vary with altitude layers. (a) shows that the intercept term α and the altitude drift coefficient β tend to stabilize with increasing altitude; (b) shows that the response coefficient γ0 of this layer is higher at low altitudes and slightly decreases at high altitudes; (c) shows the coupling weights of the lower-layer neighborhood. , As the height layer changes, (d) represents the coupling weight of the upper-layer neighborhood. , As the altitude layer changes, (c) and (d) show the variation of the neighborhood weight coefficient with altitude, generally satisfying the following trend. + Constraints.

[0061] Step 4: Overall Profile Correction and Effect Evaluation

[0062] The raw lidar data from the validation set is used to construct X(h) for each height layer according to the method in step S2. This X(h) is then input into the trained model to calculate the corrected ozone concentration. This allows us to obtain the correction value for the entire profile.

[0063] Radiosonde ozone data was used as a true reference to compare and evaluate the laser ozone data before and after correction. Based on the radiosonde ozone concentration, evaluation indicators were calculated for both the original and corrected laser ozone data. The comparative statistical results are shown in Table 1. Table 1. Comparison of correction effects of various methods using sounding ozone as the true reference.

[0064] As shown in Table 1, using sounding ozone data as the true reference, the method of this invention reduced RMSE from 32.51 ppb to 17.17 ppb, an improvement rate of 47.2%, and MAE from 25.93 ppb to 13.57 ppb, an improvement rate of 47.7%, compared with the original data before correction. Compared with the traditional hierarchical linear correction method, the method of this invention improves the overall correction effect. More importantly, it achieves joint correction of the entire profile, avoiding the problem of uneven correction effect of the entire profile caused by hierarchical independent modeling.

[0065] Figure 3 The evaluation metrics at each altitude level are shown, where (a) shows the coefficient of determination R. 2The correlation coefficients were significantly improved across all altitude layers; (b) shows a comparison of RMSE before and after correction, and (c) shows a comparison of MAE before and after correction. (b) and (c) show that RMSE and MAE were significantly reduced after correction, especially at altitudes of 4500–5000m, where the RMSE improvement rate reached 68.8% and the MAE improvement rate reached 68.5%; (d) shows the trend of RMSE and MAE improvement rates with altitude layers, showing that the improvement rate increases with altitude, indicating that the correction effect of the method of the present invention is particularly outstanding in high-altitude low signal-to-noise ratio regions.

[0066] Figure 4 To compare the scatter plots before and after correction, (a) shows that the scatter plots of the original data and the true sounding values ​​were severely dispersed before correction, deviating from the 1:1 line; (b) shows that the data after correction converged significantly towards the 1:1 line; (c) provides a visual comparison of the degree of deviation from the 1:1 reference line before and after correction; and (d) shows that the distribution of the absolute deviation shifted significantly to the left (closer to zero) after correction, proving that the model effectively suppressed random noise and corrected systematic bias.

[0067] Further analysis of the correction effect at different altitude levels reveals a more significant improvement in the high-altitude low signal-to-noise ratio region. At altitudes of 4500-5000m, the method of this invention achieves an RMSE improvement rate of 68.8% and a MAE improvement rate of 68.5%. It is particularly noteworthy that the traditional hierarchical linear correction method, with only three valid samples at each altitude level in this region, independently fits two parameters (intercept and slope). The resulting RMSE improvement rate (81.6%) and MAE improvement rate (86.7%) are essentially due to severe overfitting (R² exceeds 0.99 in approximately 37% of altitude levels), and lack true generalization ability. In contrast, the method of this invention achieves robust correction results with a significantly fewer total number of parameters than traditional methods through joint modeling of the entire profile. More importantly, it achieves a balanced correction effect across all altitude levels. This balance not only effectively improves the accuracy in the high-altitude weak signal region but also significantly alleviates the accumulation of correction bias and overfitting problems caused by independent modeling in the mid- and low-altitude layers, making the overall profile shape more consistent with the laws of atmospheric physical evolution. Meanwhile, joint modeling significantly reduces the total number of model parameters and improves computational efficiency and generalization ability.

[0068] Step 5: Model Dynamic Update Mechanism

[0069] To adapt to instrument performance drift and seasonal changes, this embodiment sets a rolling time window (e.g., one month). Each month of newly accumulated radiosonde matching data automatically incorporates new spatiotemporal matching sample pairs and automatically triggers a model retraining process to update model parameters. If a significant abrupt change (more than twice the standard deviation) is detected in model parameters (such as the β-height drift coefficient), an instrument performance warning is issued, prompting maintenance personnel to check the laser energy or optical path system.

[0070] Example 2

[0071] This embodiment provides an electronic device, the electronic device comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the method steps of the present invention as described in the above embodiments.

[0072] An electronic device according to embodiments of the present disclosure includes a memory and a processor. The memory is used to store non-transitory computer-readable instructions. Specifically, the memory 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, for example, include random access memory (RAM) and / or cache memory. The non-volatile memory may, for example, include read-only memory (ROM), hard disk, flash memory, etc.

[0073] The processor may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. In one embodiment of this disclosure, the processor is used to execute computer-readable instructions stored in the memory.

[0074] Those skilled in the art will understand that, in order to solve the technical problem of how to achieve a good user experience, this embodiment may also include well-known structures such as communication buses and interfaces, and these well-known structures should also be included within the protection scope of this disclosure.

[0075] For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0076] Example 3

[0077] This embodiment provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to execute the method of the present invention described in the above embodiment.

[0078] A computer-readable storage medium according to embodiments of the present disclosure stores non-transitory computer-readable instructions. When these non-transitory computer-readable instructions are executed by a processor, all or part of the steps of the methods described in the foregoing embodiments of the present disclosure are performed.

[0079] The aforementioned computer-readable storage media include, but are not limited to: optical storage media (e.g., CD-ROM and DVD), magneto-optical storage media (e.g., MO), magnetic storage media (e.g., magnetic tape or portable hard drive), media with built-in rewritable non-volatile memory (e.g., memory card), and media with built-in ROM (e.g., ROM cartridge).

[0080] The various embodiments of the present invention have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments.

Claims

1. An adaptive correction method for the full profile of a laser ozone radar that takes into account vertical spatial correlation, characterized in that, include: S1. Acquire spatiotemporally matched lidar ozone radar observation profile data and high-precision reference ozone profile data within historical time periods, and remove invalid data, outliers and excessively large data deviations to form sample pairs corresponding to each altitude layer. Each sample pair contains: the original lidar ozone concentration corresponding to an altitude layer, the reference ozone concentration corresponding to that altitude layer, and the altitude information corresponding to that altitude layer. S2. For each height layer, extract the original laser ozone concentration of that height layer, and extract the original laser ozone concentrations of the k adjacent lower and k adjacent upper height layers in the vertical direction. Combine the extracted original laser ozone concentrations with the height information of that height layer to form a vertical spatial correlation feature vector. Specifically, for the bottom boundary layer of the profile, only extract its k upper neighbor features, and for the top boundary layer of the profile, only extract its k lower neighbor features. S3. Establish a multiple regression model with the reference ozone concentration corresponding to the altitude layer as the dependent variable and the vertical spatial correlation feature vector corresponding to the same altitude layer as the independent variable. S4. Use the vertical spatial correlation feature vectors corresponding to each height layer and the reference ozone concentration corresponding to each height layer to form a training dataset, and train the multivariate regression model. Use the least squares method or regularized regression algorithm to solve the parameters of the multivariate regression model and obtain the optimal parameter combination. S5. For the original laser ozone radar data to be corrected, construct the vertical spatial correlation feature vector for each height layer according to the method in step S2, input it into the trained multivariate regression model, and calculate the corrected high-precision ozone vertical distribution data.

2. The method according to claim 1, characterized in that, The spatiotemporal matching described in step S1 includes time synchronization and height matching; The time synchronization is such that the time difference between the reference observation transit period and the lidar observation period does not exceed a preset time threshold. The height matching is defined as the absolute value of the difference between the height of the reference observation data and the observation height of the lidar not exceeding a preset height threshold.

3. The method according to claim 1, characterized in that, In step S2, the formula for constructing the vertical spatial correlation feature vector is expressed as follows: X(h) = [h, , , ..., , , ..., ] Where X(h) is the vertical spatial correlation feature vector of the altitude layer, and h is the altitude of that altitude layer. This represents the original laser ozone concentration at that altitude. The original laser ozone concentration in the upper and lower neighboring height layers. i =1,2,...,k, where k is the number of one-sided neighborhood layers.

4. The method according to claim 1, characterized in that, In step S3, the mathematical expression of the multiple regression model is: in, The corrected ozone concentration is represented by α; the intercept term is represented by β; the height drift coefficient is represented by γ0; and the response coefficient of this layer is represented by γ0. and These are the coupling weight coefficients for the i-th neighborhood of the lower and upper layers, respectively; i =1,2,...,k, where k is the number of one-sided neighborhood layers; The physical constraints of the above model parameters are as follows: The intercept term α is used to correct for system background bias. The height drift coefficient β is negative and is used to compensate for the underestimation of concentration caused by the attenuation of laser energy with increasing height, so as to correct the signal attenuation error that varies with height. The neighborhood coupling weight coefficient and The sum is not zero, and it is used to utilize the continuity of the vertical distribution of atmospheric ozone to perform low-pass filtering on high-frequency random noise at the current altitude layer in order to suppress random noise.

5. The method according to claim 1, characterized in that, In step S4, when the number of training samples is limited or multicollinearity exists, ridge regression or Lasso regression algorithms are used to optimize parameters and a regularization term is introduced to prevent the model from overfitting.

6. The method according to claim 1, characterized in that, It also includes a dynamic model update mechanism: setting a rolling time window, periodically incorporating new spatiotemporal matching observation data, and repeating steps S1 to S4 to recalculate model parameters to adapt to the performance drift, seasonal changes and the evolution of systematic errors caused by environmental factors of the lidar system.

7. The method according to claim 1, characterized in that, In step S1, the high-precision reference ozone profile data is derived from one or more fused data sources, including electrochemical ozone radiosonde, airborne ozone sensor, and satellite edge detection data.

8. The method according to claim 1, characterized in that, In step S2, the value of the number of unilateral neighboring layers k is dynamically adjusted based on the intensity of turbulent mixing and the vertical gradient in the atmospheric vertical structure. Within the atmospheric boundary layer, due to intense turbulent mixing, k takes values ​​of 2 to 4; In the free atmosphere, due to the large vertical gradient, k takes the value of 1 to 2.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1-8.

10. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions for causing a computer to perform the method described in any one of claims 1-8.

Images